2013-08-22 19:14:21 +02:00
|
|
|
/*
|
2014-01-09 20:12:20 +01:00
|
|
|
* Copyright (c) 2013, 2014 Conformal Systems LLC <info@conformal.com>
|
2013-08-22 19:14:21 +02:00
|
|
|
*
|
|
|
|
* Permission to use, copy, modify, and distribute this software for any
|
|
|
|
* purpose with or without fee is hereby granted, provided that the above
|
|
|
|
* copyright notice and this permission notice appear in all copies.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
|
|
|
|
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
|
|
|
|
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
|
|
|
|
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
|
|
|
|
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
|
|
|
|
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
|
|
|
|
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
|
|
|
|
*/
|
|
|
|
|
2014-05-08 21:48:42 +02:00
|
|
|
package txstore
|
2013-08-22 19:14:21 +02:00
|
|
|
|
|
|
|
import (
|
2013-09-09 20:14:57 +02:00
|
|
|
"errors"
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
"fmt"
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
"path/filepath"
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
"sort"
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
"sync"
|
2014-02-24 20:35:30 +01:00
|
|
|
"time"
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
"github.com/conformal/btcchain"
|
2014-05-23 04:16:50 +02:00
|
|
|
"github.com/conformal/btcnet"
|
2014-02-24 20:35:30 +01:00
|
|
|
"github.com/conformal/btcscript"
|
2013-11-22 19:42:25 +01:00
|
|
|
"github.com/conformal/btcutil"
|
2013-08-22 19:14:21 +02:00
|
|
|
"github.com/conformal/btcwire"
|
|
|
|
)
|
|
|
|
|
2013-11-22 19:42:25 +01:00
|
|
|
var (
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// ErrDuplicateInsert describes the error where an insert was ignored
|
|
|
|
// for being a duplicate.
|
|
|
|
ErrDuplicateInsert = errors.New("duplicate insert")
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// ErrInconsistentStore describes an error where the transaction store
|
|
|
|
// is in an inconsistent state. This error is unrecoverable, and the
|
|
|
|
// transaction store should be regenerated by a rescan.
|
|
|
|
ErrInconsistentStore = errors.New("inconsistant transaction store")
|
2013-11-22 19:42:25 +01:00
|
|
|
|
2014-02-24 20:35:30 +01:00
|
|
|
// ErrUnsupportedVersion represents an error where a serialized
|
|
|
|
// object is marked with a version that is no longer supported
|
|
|
|
// during deserialization.
|
|
|
|
ErrUnsupportedVersion = errors.New("version no longer supported")
|
2013-11-22 19:42:25 +01:00
|
|
|
)
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingValueError is a catch-all error interface for any error due to a
|
|
|
|
// missing value in the transaction store.
|
|
|
|
type MissingValueError interface {
|
|
|
|
Error() string
|
|
|
|
MissingValueError() error
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingBlockError describes an error where a block could not be found in
|
|
|
|
// the transaction store. The value is the height of the missing block.
|
|
|
|
type MissingBlockError int32
|
2014-01-06 19:35:07 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Error implements the error interface.
|
|
|
|
func (e MissingBlockError) Error() string {
|
|
|
|
return fmt.Sprintf("missing record for block %d", e)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingValueError implements the MissingValueError interface.
|
|
|
|
func (e MissingBlockError) MissingValueError() error {
|
|
|
|
return e
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingBlockTxError describes an error where a mined transaction could
|
|
|
|
// not be found in the transaction store. The value is the lookup key for
|
|
|
|
// the mined transaction.
|
|
|
|
type MissingBlockTxError BlockTxKey
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Error implements the error interface.
|
|
|
|
func (e MissingBlockTxError) Error() string {
|
|
|
|
return fmt.Sprintf("missing record for transaction at block %d index "+
|
|
|
|
"%d", e.BlockHeight, e.BlockIndex)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingValueError implements the MissingValueError interface.
|
|
|
|
func (e MissingBlockTxError) MissingValueError() error {
|
|
|
|
return e
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingDebitsError describes an error where a mined transaction does not
|
|
|
|
// contain any debiting records. The value is the lookup key for the mined
|
|
|
|
// transaction.
|
|
|
|
type MissingDebitsError BlockTxKey
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Error implements the error interface.
|
|
|
|
func (e MissingDebitsError) Error() string {
|
|
|
|
return fmt.Sprintf("missing record for debits at block %d index %d",
|
|
|
|
e.BlockHeight, e.BlockIndex)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingValueError implements the MissingValueError interface.
|
|
|
|
func (e MissingDebitsError) MissingValueError() error {
|
|
|
|
return e
|
|
|
|
}
|
2013-12-17 19:18:09 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingCreditError describes an error where a mined transaction does not
|
|
|
|
// contain a credit record at for some output. The value is the lookup key
|
|
|
|
// for the mined transaction's output.
|
|
|
|
type MissingCreditError BlockOutputKey
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Error implements the error interface.
|
|
|
|
func (e MissingCreditError) Error() string {
|
|
|
|
return fmt.Sprintf("missing record for received transaction output at "+
|
|
|
|
"block %d index %d output %d", e.BlockHeight, e.BlockIndex,
|
|
|
|
e.OutputIndex)
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// MissingValueError implements the MissingValueError interface.
|
|
|
|
func (e MissingCreditError) MissingValueError() error {
|
|
|
|
return e
|
|
|
|
}
|
2013-12-17 19:18:09 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// TxRecord is the record type for all transactions in the store. If the
|
|
|
|
// transaction is mined, BlockTxKey will be the lookup key for the transaction.
|
|
|
|
// Otherwise, the embedded BlockHeight will be -1.
|
|
|
|
type TxRecord struct {
|
|
|
|
BlockTxKey
|
|
|
|
*txRecord
|
|
|
|
s *Store
|
|
|
|
}
|
2013-08-22 19:14:21 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Block holds details about a block that contains wallet transactions.
|
|
|
|
type Block struct {
|
|
|
|
Hash btcwire.ShaHash
|
|
|
|
Time time.Time
|
|
|
|
Height int32
|
|
|
|
}
|
2013-08-22 19:14:21 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Debits is the type representing any TxRecord which debits from previous
|
|
|
|
// wallet transaction outputs.
|
|
|
|
type Debits struct {
|
|
|
|
*TxRecord
|
|
|
|
}
|
Perform smarter UTXO tracking.
This change fixes many issues with the tracking of unspent transaction
outputs. First, notifications for when UTXOs arse spent are now
requested from btcd, and when spent, will be removed from the
UtxoStore.
Second, when transactions are created, the unconfirmed (not yet in a
block) Utxo (with block height -1 and zeroed block hash) is added to
the wallet's UtxoStore. Notifications for when this UTXO is spent are
also requested from btcd. After the tx appears in a block, because
the UTXO has a pkScript to be spent by another owned wallet address, a
notification with the UTXO will be sent to btcwallet. We already
store the unconfirmed UTXO, so at this point the actual block height
and hash are filled in.
Finally, when calculating the balance, if confirmations is zero,
unconfirmed UTXOs (block height -1) will be included in the balance.
Otherwise, they are ignored.
2013-10-22 15:55:53 +02:00
|
|
|
|
2014-05-08 21:36:18 +02:00
|
|
|
// Credit is the type representing a transaction output which was spent or
|
2014-05-09 04:21:31 +02:00
|
|
|
// is still spendable by wallet. A UTXO is an unspent Credit, but not all
|
2014-05-08 21:36:18 +02:00
|
|
|
// Credits are UTXOs.
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
type Credit struct {
|
|
|
|
*TxRecord
|
|
|
|
OutputIndex uint32
|
|
|
|
}
|
2013-08-22 19:14:21 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// blockAmounts holds the immediately spendable and reward (coinbase) amounts
|
|
|
|
// as a result of all transactions in a block.
|
|
|
|
type blockAmounts struct {
|
|
|
|
Spendable btcutil.Amount
|
|
|
|
Reward btcutil.Amount
|
|
|
|
}
|
2013-09-09 19:31:37 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Store implements a transaction store for storing and managing wallet
|
|
|
|
// transactions.
|
|
|
|
type Store struct {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
// TODO: Use atomic operations for dirty so the reader lock
|
|
|
|
// doesn't need to be grabbed.
|
|
|
|
dirty bool
|
|
|
|
path string
|
|
|
|
dir string
|
|
|
|
file string
|
|
|
|
|
|
|
|
mtx sync.RWMutex
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// blocks holds wallet transaction records for each block they appear
|
|
|
|
// in. This is sorted by block height in increasing order. A separate
|
|
|
|
// map is included to lookup indexes for blocks at some height.
|
|
|
|
blocks []*blockTxCollection
|
|
|
|
blockIndexes map[int32]uint32
|
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
unspent map[btcwire.OutPoint]BlockTxKey
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
|
|
|
|
// unconfirmed holds a collection of wallet transactions that have not
|
|
|
|
// been mined into a block yet.
|
|
|
|
unconfirmed unconfirmedStore
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
|
|
|
|
// Channels to notify callers of changes to the transaction store.
|
|
|
|
// These are only created when a caller calls the appropiate
|
|
|
|
// registration method.
|
|
|
|
newCredit chan Credit
|
|
|
|
newDebits chan Debits
|
|
|
|
minedCredit chan Credit
|
|
|
|
minedDebits chan Debits
|
|
|
|
notificationLock sync.Locker
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// blockTxCollection holds a collection of wallet transactions from exactly one
|
|
|
|
// block.
|
|
|
|
type blockTxCollection struct {
|
|
|
|
// Block holds the hash, time, and height of the block.
|
|
|
|
Block
|
|
|
|
|
|
|
|
// amountDeltas is the net increase or decrease of BTC controllable by
|
|
|
|
// wallet addresses due to transactions in this block. This value only
|
|
|
|
// tracks the total amount, not amounts for individual addresses (which
|
|
|
|
// this txstore implementation is not aware of).
|
|
|
|
amountDeltas blockAmounts
|
|
|
|
|
|
|
|
// txs holds all transaction records for a block, sorted by block index.
|
|
|
|
// A separate map is included to lookup indexes for txs at some block
|
|
|
|
// index.
|
|
|
|
txs []*txRecord
|
|
|
|
txIndexes map[int]uint32
|
|
|
|
}
|
|
|
|
|
|
|
|
// unconfirmedStore stores all unconfirmed transactions managed by the Store.
|
|
|
|
type unconfirmedStore struct {
|
|
|
|
// txs contains all unconfirmed transactions, mapping from their
|
|
|
|
// transaction hash to the records.
|
|
|
|
txs map[btcwire.ShaHash]*txRecord
|
|
|
|
|
|
|
|
// spentBlockOutPoints maps from spent outputs from mined transaction
|
|
|
|
// to the unconfirmed transaction which spends it. An additional
|
|
|
|
// map is included to lookup the output key by its outpoint.
|
|
|
|
spentBlockOutPoints map[BlockOutputKey]*txRecord
|
|
|
|
spentBlockOutPointKeys map[btcwire.OutPoint]BlockOutputKey
|
|
|
|
|
|
|
|
// spentUnconfirmed maps from an unconfirmed outpoint to the unconfirmed
|
|
|
|
// transaction which spends it.
|
|
|
|
spentUnconfirmed map[btcwire.OutPoint]*txRecord
|
|
|
|
|
|
|
|
// previousOutpoints maps all previous outputs to the transaction record
|
|
|
|
// of the unconfirmed transaction which spends it. This is primarly
|
|
|
|
// designed to assist with double spend detection without iterating
|
|
|
|
// through each value of the txs map.
|
|
|
|
previousOutpoints map[btcwire.OutPoint]*txRecord
|
|
|
|
}
|
|
|
|
|
|
|
|
// BlockTxKey is a lookup key for a single mined transaction in the store.
|
|
|
|
type BlockTxKey struct {
|
|
|
|
BlockIndex int
|
|
|
|
BlockHeight int32
|
|
|
|
}
|
|
|
|
|
|
|
|
// BlockOutputKey is a lookup key for a transaction output from a block in the
|
|
|
|
// store.
|
|
|
|
type BlockOutputKey struct {
|
|
|
|
BlockTxKey
|
|
|
|
OutputIndex uint32
|
|
|
|
}
|
|
|
|
|
|
|
|
// txRecord holds all credits and debits created by a transaction's inputs and
|
|
|
|
// outputs.
|
|
|
|
type txRecord struct {
|
|
|
|
// tx is the transaction that all records in this structure reference.
|
|
|
|
tx *btcutil.Tx
|
2013-08-22 19:14:21 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// debit records the debits this transaction creates, or nil if the
|
|
|
|
// transaction does not spend any previous credits.
|
|
|
|
debits *debits
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// credits holds all records for received transaction outputs from
|
|
|
|
// this transaction.
|
|
|
|
credits []*credit
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
received time.Time
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// debits records the debits a transaction record makes from previous wallet
|
|
|
|
// transaction credits.
|
|
|
|
type debits struct {
|
|
|
|
amount btcutil.Amount
|
2014-06-18 23:46:18 +02:00
|
|
|
spends []BlockOutputKey
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// credit describes a transaction output which was or is spendable by wallet.
|
|
|
|
type credit struct {
|
|
|
|
change bool
|
|
|
|
spentBy *BlockTxKey // nil if unspent
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
2014-05-08 21:48:42 +02:00
|
|
|
// New allocates and initializes a new transaction store.
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
func New(dir string) *Store {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return &Store{
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
path: filepath.Join(dir, filename),
|
|
|
|
dir: dir,
|
|
|
|
file: filename,
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
blockIndexes: map[int32]uint32{},
|
2014-06-18 05:48:54 +02:00
|
|
|
unspent: map[btcwire.OutPoint]BlockTxKey{},
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
unconfirmed: unconfirmedStore{
|
|
|
|
txs: map[btcwire.ShaHash]*txRecord{},
|
|
|
|
spentBlockOutPoints: map[BlockOutputKey]*txRecord{},
|
|
|
|
spentBlockOutPointKeys: map[btcwire.OutPoint]BlockOutputKey{},
|
|
|
|
spentUnconfirmed: map[btcwire.OutPoint]*txRecord{},
|
|
|
|
previousOutpoints: map[btcwire.OutPoint]*txRecord{},
|
|
|
|
},
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
notificationLock: new(sync.Mutex),
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (s *Store) lookupBlock(height int32) (*blockTxCollection, error) {
|
|
|
|
if i, ok := s.blockIndexes[height]; ok {
|
|
|
|
return s.blocks[i], nil
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return nil, MissingBlockError(height)
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (s *Store) lookupBlockTx(key BlockTxKey) (*txRecord, error) {
|
|
|
|
coll, err := s.lookupBlock(key.BlockHeight)
|
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if i, ok := coll.txIndexes[key.BlockIndex]; ok {
|
|
|
|
return coll.txs[i], nil
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return nil, MissingBlockTxError{key.BlockIndex, key.BlockHeight}
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (s *Store) lookupBlockDebits(key BlockTxKey) (*debits, error) {
|
|
|
|
r, err := s.lookupBlockTx(key)
|
2014-02-28 19:03:23 +01:00
|
|
|
if err != nil {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return nil, err
|
2014-02-28 19:03:23 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if r.debits == nil {
|
|
|
|
return nil, MissingDebitsError(key)
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return r.debits, nil
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (r *txRecord) lookupBlockCredit(key BlockOutputKey) (*credit, error) {
|
|
|
|
switch {
|
|
|
|
case len(r.credits) <= int(key.OutputIndex):
|
|
|
|
fallthrough
|
|
|
|
case r.credits[key.OutputIndex] == nil:
|
|
|
|
return nil, MissingCreditError(key)
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return r.credits[key.OutputIndex], nil
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (s *Store) lookupBlockCredit(key BlockOutputKey) (*credit, error) {
|
|
|
|
txRecord, err := s.lookupBlockTx(key.BlockTxKey)
|
2014-02-28 19:03:23 +01:00
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return txRecord.lookupBlockCredit(key)
|
2013-12-17 19:18:09 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (s *Store) blockCollectionForInserts(block *Block) *blockTxCollection {
|
|
|
|
b, err := s.lookupBlock(block.Height)
|
|
|
|
switch e := err.(type) {
|
|
|
|
case MissingBlockError:
|
|
|
|
b = &blockTxCollection{
|
|
|
|
Block: *block,
|
|
|
|
txIndexes: map[int]uint32{},
|
|
|
|
}
|
|
|
|
|
|
|
|
// If this new block cannot be appended to the end of the blocks
|
|
|
|
// slice (which would disobey ordering blocks by their height),
|
|
|
|
// reslice and update the store's map of block heights to block
|
|
|
|
// slice indexes.
|
|
|
|
if len(s.blocks) > 0 && s.blocks[len(s.blocks)-1].Height > block.Height {
|
|
|
|
i := uint32(len(s.blocks))
|
|
|
|
for i != 0 && s.blocks[i-1].Height >= block.Height {
|
|
|
|
i--
|
|
|
|
}
|
|
|
|
detached := s.blocks[i:]
|
|
|
|
s.blocks = append(s.blocks[:i], b)
|
|
|
|
s.blockIndexes[b.Height] = i
|
|
|
|
for i, b := range detached {
|
|
|
|
newIndex := uint32(i + len(s.blocks))
|
|
|
|
s.blockIndexes[b.Height] = newIndex
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
s.blocks = append(s.blocks, detached...)
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
} else {
|
|
|
|
s.blockIndexes[int32(e)] = uint32(len(s.blocks))
|
|
|
|
s.blocks = append(s.blocks, b)
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return b
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (c *blockTxCollection) lookupTxRecord(blockIndex int) (*txRecord, uint32, error) {
|
|
|
|
if i, ok := c.txIndexes[blockIndex]; ok {
|
|
|
|
return c.txs[i], i, nil
|
|
|
|
}
|
|
|
|
return nil, 0, MissingBlockTxError{blockIndex, c.Block.Height}
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (c *blockTxCollection) txRecordForInserts(tx *btcutil.Tx) *txRecord {
|
|
|
|
if i, ok := c.txIndexes[tx.Index()]; ok {
|
|
|
|
return c.txs[i]
|
|
|
|
}
|
2014-06-20 01:11:47 +02:00
|
|
|
|
|
|
|
log.Infof("Inserting transaction %v from block %d", tx.Sha(), c.Height)
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
record := &txRecord{tx: tx}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If this new transaction record cannot be appended to the end of the
|
|
|
|
// txs slice (which would disobey ordering transactions by their block
|
|
|
|
// index), reslice and update the block's map of block indexes to txs
|
|
|
|
// slice indexes.
|
|
|
|
if len(c.txs) > 0 && c.txs[len(c.txs)-1].Tx().Index() > tx.Index() {
|
|
|
|
i := uint32(len(c.txs))
|
|
|
|
for i != 0 && c.txs[i-1].Tx().Index() >= tx.Index() {
|
|
|
|
i--
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
detached := c.txs[i:]
|
|
|
|
c.txs = append(c.txs[:i], record)
|
|
|
|
c.txIndexes[tx.Index()] = i
|
|
|
|
for i, r := range detached {
|
|
|
|
newIndex := uint32(i + len(c.txs))
|
|
|
|
c.txIndexes[r.Tx().Index()] = newIndex
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
c.txs = append(c.txs, detached...)
|
|
|
|
} else {
|
|
|
|
c.txIndexes[tx.Index()] = uint32(len(c.txs))
|
|
|
|
c.txs = append(c.txs, record)
|
2013-12-17 19:18:09 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return record
|
|
|
|
}
|
2013-12-17 19:18:09 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (s *Store) blockTxRecordForInserts(tx *btcutil.Tx, block *Block) *txRecord {
|
|
|
|
return s.blockCollectionForInserts(block).txRecordForInserts(tx)
|
2013-12-17 19:18:09 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (u *unconfirmedStore) txRecordForInserts(tx *btcutil.Tx) *txRecord {
|
|
|
|
r, ok := u.txs[*tx.Sha()]
|
|
|
|
if !ok {
|
2014-06-20 01:11:47 +02:00
|
|
|
log.Infof("Inserting unconfirmed transaction %v", tx.Sha())
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
r = &txRecord{tx: tx}
|
|
|
|
u.txs[*tx.Sha()] = r
|
|
|
|
for _, input := range r.Tx().MsgTx().TxIn {
|
|
|
|
u.previousOutpoints[input.PreviousOutpoint] = r
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return r
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (s *Store) moveMinedTx(r *txRecord, block *Block) error {
|
2014-06-20 01:11:47 +02:00
|
|
|
log.Infof("Marking unconfirmed transaction %v mined in block %d",
|
|
|
|
r.tx.Sha(), block.Height)
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
delete(s.unconfirmed.txs, *r.Tx().Sha())
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Find collection and insert records. Error out if there are records
|
|
|
|
// saved for this block and index.
|
|
|
|
key := BlockTxKey{r.Tx().Index(), block.Height}
|
|
|
|
b := s.blockCollectionForInserts(block)
|
|
|
|
txIndex := uint32(len(b.txs))
|
|
|
|
b.txIndexes[key.BlockIndex] = txIndex
|
|
|
|
b.txs = append(b.txs, r)
|
2013-08-22 19:14:21 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for _, input := range r.Tx().MsgTx().TxIn {
|
|
|
|
delete(s.unconfirmed.previousOutpoints, input.PreviousOutpoint)
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// For all mined transactions with credits spent by this
|
|
|
|
// transaction, remove them from the spentBlockOutPoints map
|
|
|
|
// (signifying that there is no longer an unconfirmed
|
|
|
|
// transaction which spending that credit), and update the
|
|
|
|
// credit's spent by tracking with the block key of the
|
|
|
|
// newly-mined transaction.
|
|
|
|
prev, ok := s.unconfirmed.spentBlockOutPointKeys[input.PreviousOutpoint]
|
|
|
|
if !ok {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
delete(s.unconfirmed.spentBlockOutPointKeys, input.PreviousOutpoint)
|
|
|
|
delete(s.unconfirmed.spentBlockOutPoints, prev)
|
|
|
|
rr, err := s.lookupBlockTx(prev.BlockTxKey)
|
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
rr.credits[prev.OutputIndex].spentBy = &key
|
|
|
|
// debits should already be non-nil
|
2014-06-18 23:46:18 +02:00
|
|
|
r.debits.spends = append(r.debits.spends, prev)
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
if r.debits != nil {
|
|
|
|
d := Debits{&TxRecord{key, r, s}}
|
|
|
|
s.notifyMinedDebits(d)
|
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
|
|
|
|
// For each credit in r, if the credit is spent by another unconfirmed
|
|
|
|
// transaction, move the spending transaction from spentUnconfirmed
|
|
|
|
// (which signifies a transaction spending another unconfirmed tx) to
|
|
|
|
// spentBlockTxs (which signifies an unconfirmed transaction spending a
|
|
|
|
// confirmed tx) and modify the mined transaction's record to refer to
|
|
|
|
// the credit being spent by an unconfirmed transaction.
|
|
|
|
//
|
|
|
|
// If the credit is not spent, modify the store's unspent bookkeeping
|
|
|
|
// maps to include the credit and increment the amount deltas by the
|
|
|
|
// credit's value.
|
2014-06-18 23:46:18 +02:00
|
|
|
op := btcwire.OutPoint{Hash: *r.Tx().Sha()}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for i, credit := range r.credits {
|
|
|
|
if credit == nil {
|
|
|
|
continue
|
|
|
|
}
|
2014-06-18 23:46:18 +02:00
|
|
|
op.Index = uint32(i)
|
|
|
|
outputKey := BlockOutputKey{key, op.Index}
|
|
|
|
if rr, ok := s.unconfirmed.spentUnconfirmed[op]; ok {
|
|
|
|
delete(s.unconfirmed.spentUnconfirmed, op)
|
|
|
|
s.unconfirmed.spentBlockOutPointKeys[op] = outputKey
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
s.unconfirmed.spentBlockOutPoints[outputKey] = rr
|
|
|
|
credit.spentBy = &BlockTxKey{BlockHeight: -1}
|
|
|
|
} else if credit.spentBy == nil {
|
2014-06-18 05:48:54 +02:00
|
|
|
// Mark outpoint unspent.
|
2014-06-18 23:46:18 +02:00
|
|
|
s.unspent[op] = key
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
|
|
|
|
// Increment spendable amount delta as a result of
|
|
|
|
// moving this credit to this block.
|
|
|
|
value := r.Tx().MsgTx().TxOut[i].Value
|
|
|
|
b.amountDeltas.Spendable += btcutil.Amount(value)
|
|
|
|
}
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
|
|
|
|
c := Credit{&TxRecord{key, r, s}, op.Index}
|
|
|
|
s.notifyMinedCredit(c)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If this moved transaction debits from any previous credits, decrement
|
|
|
|
// the amount deltas by the total input amount. Because this
|
|
|
|
// transaction was moved from the unconfirmed transaction set, this can
|
|
|
|
// never be a coinbase.
|
|
|
|
if r.debits != nil {
|
|
|
|
b.amountDeltas.Spendable -= r.debits.amount
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return nil
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// InsertTx records a transaction as belonging to a wallet's transaction
|
|
|
|
// history. If block is nil, the transaction is considered unspent, and the
|
|
|
|
// transaction's index must be unset. Otherwise, the transaction index must be
|
|
|
|
// set if a non-nil block is set.
|
|
|
|
//
|
|
|
|
// The transaction record is returned. Credits and debits may be added to the
|
|
|
|
// transaction by calling methods on the TxRecord.
|
|
|
|
func (s *Store) InsertTx(tx *btcutil.Tx, block *Block) (*TxRecord, error) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
s.mtx.Lock()
|
|
|
|
defer s.mtx.Unlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// The receive time will be the earlier of now and the block time
|
|
|
|
// (if any).
|
|
|
|
received := time.Now()
|
|
|
|
|
|
|
|
// Verify that the index of the transaction within the block is
|
|
|
|
// set if a block is set, and unset if there is no block.
|
|
|
|
index := tx.Index()
|
|
|
|
switch {
|
|
|
|
case index == btcutil.TxIndexUnknown && block != nil:
|
|
|
|
return nil, errors.New("transaction block index unset")
|
|
|
|
case index != btcutil.TxIndexUnknown && block == nil:
|
|
|
|
return nil, errors.New("transaction block index set")
|
|
|
|
}
|
|
|
|
|
|
|
|
// Simply create or return the transaction record if this transaction
|
|
|
|
// is unconfirmed.
|
|
|
|
if block == nil {
|
|
|
|
r := s.unconfirmed.txRecordForInserts(tx)
|
|
|
|
r.received = received
|
2014-06-18 23:46:18 +02:00
|
|
|
return &TxRecord{BlockTxKey{BlockHeight: -1}, r, s}, nil
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// Check if block records already exist for this tx. If so,
|
|
|
|
// we're done.
|
|
|
|
key := BlockTxKey{index, block.Height}
|
|
|
|
record, err := s.lookupBlockTx(key)
|
|
|
|
switch err.(type) {
|
|
|
|
case MissingValueError:
|
|
|
|
// handle it later
|
|
|
|
case nil:
|
|
|
|
// Verify that the txs actually match.
|
|
|
|
if *record.tx.Sha() != *tx.Sha() {
|
|
|
|
return nil, ErrInconsistentStore
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return &TxRecord{key, record, s}, nil
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If the exact tx (not a double spend) is already included but
|
|
|
|
// unconfirmed, move it to a block.
|
|
|
|
if r, ok := s.unconfirmed.txs[*tx.Sha()]; ok {
|
|
|
|
r.Tx().SetIndex(tx.Index())
|
|
|
|
if err := s.moveMinedTx(r, block); err != nil {
|
|
|
|
return nil, err
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return &TxRecord{key, r, s}, nil
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If this transaction is not already saved unconfirmed, remove all
|
|
|
|
// unconfirmed transactions that are now invalidated due to being a
|
|
|
|
// double spend. This also handles killing unconfirmed transaction
|
|
|
|
// spend chains if any other unconfirmed transactions spend outputs
|
|
|
|
// of the removed double spend.
|
|
|
|
if err := s.removeDoubleSpends(tx); err != nil {
|
|
|
|
return nil, err
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
r := s.blockTxRecordForInserts(tx, block)
|
|
|
|
if r.received.IsZero() {
|
|
|
|
if !block.Time.IsZero() && block.Time.Before(received) {
|
|
|
|
received = block.Time
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
r.received = received
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return &TxRecord{key, r, s}, nil
|
2013-09-09 20:14:57 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Received returns the earliest known time the transaction was received by.
|
|
|
|
func (t *TxRecord) Received() time.Time {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return t.received
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-12-17 19:18:09 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Block returns the block details for a transaction. If the transaction is
|
|
|
|
// unmined, both the block and returned error are nil.
|
|
|
|
func (t *TxRecord) Block() (*Block, error) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
coll, err := t.s.lookupBlock(t.BlockHeight)
|
|
|
|
if err != nil {
|
|
|
|
if err == MissingBlockError(-1) {
|
|
|
|
return nil, nil
|
|
|
|
}
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
return &coll.Block, nil
|
2013-12-17 19:18:09 +01:00
|
|
|
}
|
|
|
|
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
// AddDebits marks a transaction record as having debited from previous wallet
|
|
|
|
// credits.
|
|
|
|
func (t *TxRecord) AddDebits() (Debits, error) {
|
|
|
|
t.s.mtx.Lock()
|
|
|
|
defer t.s.mtx.Unlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if t.debits == nil {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
spent, err := t.s.findPreviousCredits(t.Tx())
|
|
|
|
if err != nil {
|
|
|
|
return Debits{}, err
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
|
|
|
debitAmount, err := t.s.markOutputsSpent(spent, t)
|
2014-02-24 20:35:30 +01:00
|
|
|
if err != nil {
|
2014-06-18 07:16:08 +02:00
|
|
|
return Debits{}, err
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
|
|
|
|
prevOutputKeys := make([]BlockOutputKey, len(spent))
|
|
|
|
for i, c := range spent {
|
|
|
|
prevOutputKeys[i] = c.outputKey()
|
|
|
|
}
|
|
|
|
|
|
|
|
t.debits = &debits{amount: debitAmount, spends: prevOutputKeys}
|
2014-06-20 01:11:47 +02:00
|
|
|
|
|
|
|
log.Debugf("Transaction %v spends %d previously-unspent "+
|
|
|
|
"%s totaling %v", t.tx.Sha(), len(spent),
|
|
|
|
pickNoun(len(spent), "output", "outputs"), debitAmount)
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
|
|
|
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
d := Debits{t}
|
|
|
|
t.s.notifyNewDebits(d)
|
|
|
|
return d, nil
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2013-10-15 16:05:51 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// findPreviousCredits searches for all unspent credits that make up the inputs
|
2014-06-18 05:48:54 +02:00
|
|
|
// for tx.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (s *Store) findPreviousCredits(tx *btcutil.Tx) ([]Credit, error) {
|
2014-06-18 05:48:54 +02:00
|
|
|
type createdCredit struct {
|
2014-06-18 07:16:08 +02:00
|
|
|
credit Credit
|
2014-06-18 05:48:54 +02:00
|
|
|
err error
|
2013-10-15 16:05:51 +02:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
inputs := tx.MsgTx().TxIn
|
|
|
|
creditChans := make([]chan createdCredit, len(inputs))
|
|
|
|
for i, txIn := range inputs {
|
|
|
|
creditChans[i] = make(chan createdCredit)
|
|
|
|
go func(i int, op btcwire.OutPoint) {
|
|
|
|
key, ok := s.unspent[op]
|
|
|
|
if !ok {
|
2014-06-19 18:39:43 +02:00
|
|
|
// Does this input spend an unconfirmed output?
|
|
|
|
r, ok := s.unconfirmed.txs[op.Hash]
|
|
|
|
switch {
|
|
|
|
// Not an unconfirmed tx.
|
|
|
|
case !ok:
|
|
|
|
fallthrough
|
|
|
|
// Output isn't a credit.
|
|
|
|
case len(r.credits) <= int(op.Index):
|
|
|
|
fallthrough
|
|
|
|
// Output isn't a credit.
|
|
|
|
case r.credits[op.Index] == nil:
|
|
|
|
fallthrough
|
|
|
|
// Credit already spent.
|
|
|
|
case s.unconfirmed.spentUnconfirmed[op] != nil:
|
|
|
|
close(creditChans[i])
|
|
|
|
return
|
|
|
|
}
|
|
|
|
t := &TxRecord{BlockTxKey{BlockHeight: -1}, r, s}
|
|
|
|
c := Credit{t, op.Index}
|
|
|
|
creditChans[i] <- createdCredit{credit: c}
|
2014-06-18 05:48:54 +02:00
|
|
|
return
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-06-18 05:48:54 +02:00
|
|
|
r, err := s.lookupBlockTx(key)
|
|
|
|
if err != nil {
|
|
|
|
creditChans[i] <- createdCredit{err: err}
|
|
|
|
return
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-06-18 05:48:54 +02:00
|
|
|
t := &TxRecord{key, r, s}
|
2014-06-18 07:16:08 +02:00
|
|
|
c := Credit{t, op.Index}
|
2014-06-18 05:48:54 +02:00
|
|
|
creditChans[i] <- createdCredit{credit: c}
|
|
|
|
}(i, txIn.PreviousOutpoint)
|
|
|
|
}
|
2014-06-18 07:16:08 +02:00
|
|
|
spent := make([]Credit, 0, len(inputs))
|
2014-06-18 05:48:54 +02:00
|
|
|
for _, c := range creditChans {
|
|
|
|
cc, ok := <-c
|
|
|
|
if !ok {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
if cc.err != nil {
|
|
|
|
return nil, cc.err
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
2014-06-18 05:48:54 +02:00
|
|
|
spent = append(spent, cc.credit)
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return spent, nil
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// markOutputsSpent marks each previous credit spent by t as spent. The total
|
|
|
|
// input of all spent previous outputs is returned.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (s *Store) markOutputsSpent(spent []Credit, t *TxRecord) (btcutil.Amount, error) {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
var a btcutil.Amount
|
|
|
|
for _, prev := range spent {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
op := prev.outPoint()
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
switch prev.BlockHeight {
|
|
|
|
case -1: // unconfirmed
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
op := prev.outPoint()
|
|
|
|
switch prev.BlockHeight {
|
|
|
|
case -1: // unconfirmed
|
|
|
|
t.s.unconfirmed.spentUnconfirmed[*op] = t.txRecord
|
|
|
|
default:
|
|
|
|
key := prev.outputKey()
|
|
|
|
t.s.unconfirmed.spentBlockOutPointKeys[*op] = key
|
|
|
|
t.s.unconfirmed.spentBlockOutPoints[key] = t.txRecord
|
|
|
|
}
|
2013-09-09 19:31:37 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
default:
|
2014-06-18 05:48:54 +02:00
|
|
|
// Update spent info.
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
credit := prev.txRecord.credits[prev.OutputIndex]
|
|
|
|
if credit.spentBy != nil {
|
|
|
|
if *credit.spentBy == t.BlockTxKey {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
return 0, ErrInconsistentStore
|
|
|
|
}
|
|
|
|
credit.spentBy = &t.BlockTxKey
|
2014-06-18 05:48:54 +02:00
|
|
|
delete(s.unspent, *op)
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if t.BlockHeight == -1 { // unconfirmed
|
|
|
|
key := prev.outputKey()
|
2014-06-18 23:46:18 +02:00
|
|
|
s.unconfirmed.spentBlockOutPointKeys[*op] = key
|
|
|
|
s.unconfirmed.spentBlockOutPoints[key] = t.txRecord
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2013-09-09 19:31:37 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Increment total debited amount.
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
a += prev.amount()
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If t refers to a mined transaction, update its block's amount deltas
|
|
|
|
// by the total debited amount.
|
|
|
|
if t.BlockHeight != -1 {
|
|
|
|
b, err := s.lookupBlock(t.BlockHeight)
|
2014-02-24 20:35:30 +01:00
|
|
|
if err != nil {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return 0, err
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
b.amountDeltas.Spendable -= a
|
2013-09-09 19:31:37 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return a, nil
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
2014-06-18 23:46:18 +02:00
|
|
|
func (r *txRecord) setCredit(index uint32, change bool, tx *btcutil.Tx) error {
|
|
|
|
if r.credits == nil {
|
|
|
|
r.credits = make([]*credit, 0, len(tx.MsgTx().TxOut))
|
|
|
|
}
|
|
|
|
for i := uint32(len(r.credits)); i <= index; i++ {
|
|
|
|
r.credits = append(r.credits, nil)
|
|
|
|
}
|
|
|
|
if r.credits[index] != nil {
|
|
|
|
if *r.tx.Sha() == *tx.Sha() {
|
|
|
|
return ErrDuplicateInsert
|
|
|
|
}
|
|
|
|
return ErrInconsistentStore
|
|
|
|
}
|
|
|
|
r.credits[index] = &credit{change: change}
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// AddCredit marks the transaction record as containing a transaction output
|
|
|
|
// spendable by wallet. The output is added unspent, and is marked spent
|
|
|
|
// when a new transaction spending the output is inserted into the store.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (t *TxRecord) AddCredit(index uint32, change bool) (Credit, error) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.Lock()
|
|
|
|
defer t.s.mtx.Unlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if len(t.tx.MsgTx().TxOut) <= int(index) {
|
2014-06-18 07:16:08 +02:00
|
|
|
return Credit{}, errors.New("transaction output does not exist")
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
2014-06-18 23:46:18 +02:00
|
|
|
if err := t.txRecord.setCredit(index, change, t.tx); err != nil {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if err == ErrDuplicateInsert {
|
2014-06-18 07:16:08 +02:00
|
|
|
return Credit{t, index}, nil
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-06-18 07:16:08 +02:00
|
|
|
return Credit{}, err
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
2014-06-20 01:11:47 +02:00
|
|
|
txOutAmt := btcutil.Amount(t.tx.MsgTx().TxOut[index].Value)
|
|
|
|
log.Debugf("Marking transaction %v output %d (%v) spendable",
|
|
|
|
t.tx.Sha(), index, txOutAmt)
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
switch t.BlockHeight {
|
|
|
|
case -1: // unconfirmed
|
|
|
|
default:
|
|
|
|
b, err := t.s.lookupBlock(t.BlockHeight)
|
|
|
|
if err != nil {
|
2014-06-18 07:16:08 +02:00
|
|
|
return Credit{}, err
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// New outputs are added unspent.
|
2014-06-18 05:48:54 +02:00
|
|
|
op := btcwire.OutPoint{Hash: *t.tx.Sha(), Index: index}
|
|
|
|
t.s.unspent[op] = t.BlockTxKey
|
2014-06-20 01:11:47 +02:00
|
|
|
switch t.tx.Index() {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
case 0: // Coinbase
|
2014-06-20 01:11:47 +02:00
|
|
|
b.amountDeltas.Reward += txOutAmt
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
default:
|
2014-06-20 01:11:47 +02:00
|
|
|
b.amountDeltas.Spendable += txOutAmt
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
c := Credit{t, index}
|
|
|
|
t.s.notifyNewCredit(c)
|
|
|
|
return c, nil
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Rollback removes all blocks at height onwards, moving any transactions within
|
|
|
|
// each block to the unconfirmed pool.
|
|
|
|
func (s *Store) Rollback(height int32) error {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
s.mtx.Lock()
|
|
|
|
defer s.mtx.Unlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
i := len(s.blocks)
|
|
|
|
for i != 0 && s.blocks[i-1].Height >= height {
|
|
|
|
i--
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
detached := s.blocks[i:]
|
|
|
|
s.blocks = s.blocks[:i]
|
|
|
|
for _, b := range detached {
|
2014-06-20 15:35:14 +02:00
|
|
|
movedTxs := len(b.txs)
|
|
|
|
// Don't include coinbase transaction with number of moved txs.
|
|
|
|
// There should always be at least one tx in a block collection,
|
|
|
|
// and if there is a coinbase, it would be at index 0.
|
|
|
|
if b.txs[0].tx.Index() == 0 {
|
|
|
|
movedTxs--
|
|
|
|
}
|
|
|
|
log.Infof("Rolling back block %d (%d transactions marked "+
|
|
|
|
"unconfirmed)", b.Height, movedTxs)
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
delete(s.blockIndexes, b.Block.Height)
|
|
|
|
for _, r := range b.txs {
|
|
|
|
oldTxIndex := r.Tx().Index()
|
2013-08-22 19:14:21 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If the removed transaction is a coinbase, do not move
|
|
|
|
// it to unconfirmed.
|
|
|
|
if oldTxIndex == 0 {
|
|
|
|
continue
|
|
|
|
}
|
2013-08-22 19:14:21 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
r.Tx().SetIndex(btcutil.TxIndexUnknown)
|
|
|
|
s.unconfirmed.txs[*r.Tx().Sha()] = r
|
|
|
|
for _, input := range r.Tx().MsgTx().TxIn {
|
|
|
|
op := input.PreviousOutpoint
|
|
|
|
s.unconfirmed.previousOutpoints[op] = r
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
// For each detached spent credit, remove from the
|
|
|
|
// store's unspent map, and lookup the spender and
|
|
|
|
// modify its debit record to reference spending an
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// unconfirmed transaction.
|
|
|
|
for outIdx, credit := range r.credits {
|
|
|
|
if credit == nil {
|
|
|
|
continue
|
|
|
|
}
|
2014-06-18 05:48:54 +02:00
|
|
|
|
|
|
|
op := btcwire.OutPoint{
|
|
|
|
Hash: *r.Tx().Sha(),
|
|
|
|
Index: uint32(outIdx),
|
|
|
|
}
|
|
|
|
delete(s.unspent, op)
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
spenderKey := credit.spentBy
|
|
|
|
if spenderKey == nil {
|
|
|
|
continue
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
prev := BlockOutputKey{
|
|
|
|
BlockTxKey: BlockTxKey{
|
|
|
|
BlockIndex: oldTxIndex,
|
|
|
|
BlockHeight: b.Height,
|
|
|
|
},
|
|
|
|
OutputIndex: uint32(outIdx),
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Lookup txRecord of the spending transaction. Spent
|
|
|
|
// tracking differs slightly depending on whether the
|
|
|
|
// spender is confirmed or not.
|
|
|
|
switch spenderKey.BlockHeight {
|
|
|
|
case -1: // unconfirmed
|
|
|
|
spender, ok := s.unconfirmed.spentBlockOutPoints[prev]
|
|
|
|
if !ok {
|
|
|
|
return ErrInconsistentStore
|
|
|
|
}
|
|
|
|
|
|
|
|
// Swap the maps the spender is saved in.
|
|
|
|
delete(s.unconfirmed.spentBlockOutPointKeys, op)
|
|
|
|
delete(s.unconfirmed.spentBlockOutPoints, prev)
|
|
|
|
s.unconfirmed.spentUnconfirmed[op] = spender
|
|
|
|
|
|
|
|
default:
|
|
|
|
spender, err := s.lookupBlockTx(*spenderKey)
|
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
|
|
|
|
if spender.debits == nil {
|
|
|
|
return MissingDebitsError(*spenderKey)
|
|
|
|
}
|
|
|
|
|
|
|
|
current := BlockOutputKey{
|
|
|
|
BlockTxKey: BlockTxKey{BlockHeight: -1},
|
|
|
|
}
|
2014-06-18 23:46:18 +02:00
|
|
|
err = spender.swapDebits(prev, current)
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// If this transaction debits any previous credits,
|
|
|
|
// modify each previous credit to mark it as spent
|
|
|
|
// by an unconfirmed tx.
|
|
|
|
if r.debits != nil {
|
|
|
|
for _, prev := range r.debits.spends {
|
|
|
|
rr, err := s.lookupBlockTx(prev.BlockTxKey)
|
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
2014-06-18 23:46:18 +02:00
|
|
|
c, err := rr.lookupBlockCredit(prev)
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
op := btcwire.OutPoint{
|
|
|
|
Hash: *rr.Tx().Sha(),
|
2014-06-16 23:25:04 +02:00
|
|
|
Index: prev.OutputIndex,
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-06-18 23:46:18 +02:00
|
|
|
s.unconfirmed.spentBlockOutPointKeys[op] = prev
|
|
|
|
s.unconfirmed.spentBlockOutPoints[prev] = r
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
c.spentBy = &BlockTxKey{BlockHeight: -1}
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Debit tracking for unconfirmed transactions is
|
|
|
|
// done by the unconfirmed store's maps, and not
|
|
|
|
// in the txRecord itself.
|
|
|
|
r.debits.spends = nil
|
|
|
|
}
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
|
|
|
return nil
|
|
|
|
}
|
2014-04-09 05:04:10 +02:00
|
|
|
|
2014-06-18 23:46:18 +02:00
|
|
|
func (r *txRecord) swapDebits(previous, current BlockOutputKey) error {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for i, outputKey := range r.debits.spends {
|
2014-06-18 23:46:18 +02:00
|
|
|
if outputKey == previous {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
r.debits.spends[i] = current
|
|
|
|
return nil
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
2014-06-18 23:46:18 +02:00
|
|
|
return MissingCreditError(previous)
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// UnminedDebitTxs returns the underlying transactions for all wallet
|
|
|
|
// transactions which debit from previous outputs and are not known to have
|
|
|
|
// been mined in a block.
|
|
|
|
func (s *Store) UnminedDebitTxs() []*btcutil.Tx {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
s.mtx.RLock()
|
|
|
|
defer s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
unmined := make([]*btcutil.Tx, 0, len(s.unconfirmed.txs))
|
|
|
|
for _, r := range s.unconfirmed.spentBlockOutPoints {
|
|
|
|
unmined = append(unmined, r.Tx())
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for _, r := range s.unconfirmed.spentUnconfirmed {
|
|
|
|
unmined = append(unmined, r.Tx())
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return unmined
|
|
|
|
}
|
2013-12-17 19:18:09 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// removeDoubleSpends checks for any unconfirmed transactions which would
|
|
|
|
// introduce a double spend if tx was added to the store (either as a confirmed
|
|
|
|
// or unconfirmed transaction). If one is found, it and all transactions which
|
|
|
|
// spends its outputs (if any) are removed, and all previous inputs for any
|
|
|
|
// removed transactions are set to unspent.
|
|
|
|
func (s *Store) removeDoubleSpends(tx *btcutil.Tx) error {
|
|
|
|
if ds := s.unconfirmed.findDoubleSpend(tx); ds != nil {
|
2014-06-20 01:11:47 +02:00
|
|
|
log.Debugf("Removing double spending transaction %v", ds.tx.Sha())
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return s.removeConflict(ds)
|
2013-12-17 19:18:09 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return nil
|
2013-12-17 19:18:09 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (u *unconfirmedStore) findDoubleSpend(tx *btcutil.Tx) *txRecord {
|
|
|
|
for _, input := range tx.MsgTx().TxIn {
|
|
|
|
if r, ok := u.previousOutpoints[input.PreviousOutpoint]; ok {
|
|
|
|
return r
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return nil
|
|
|
|
}
|
2013-09-09 20:14:57 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// removeConflict removes an unconfirmed transaction record and all spend chains
|
|
|
|
// deriving from it from the store. This is designed to remove transactions
|
|
|
|
// that would otherwise result in double spend conflicts if left in the store.
|
|
|
|
// All not-removed credits spent by removed transactions are set unspent.
|
|
|
|
func (s *Store) removeConflict(r *txRecord) error {
|
|
|
|
u := &s.unconfirmed
|
2013-09-09 20:14:57 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If this transaction contains any spent credits (which must be spent by
|
|
|
|
// other unconfirmed transactions), recursively remove each spender.
|
|
|
|
for i, credit := range r.credits {
|
|
|
|
if credit == nil || credit.spentBy == nil {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
op := btcwire.NewOutPoint(r.Tx().Sha(), uint32(i))
|
|
|
|
nextSpender, ok := u.spentUnconfirmed[*op]
|
|
|
|
if !ok {
|
|
|
|
return ErrInconsistentStore
|
|
|
|
}
|
2014-06-20 01:11:47 +02:00
|
|
|
log.Debugf("Transaction %v is part of a removed double spend "+
|
|
|
|
"chain -- removing as well", nextSpender.tx.Sha())
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if err := s.removeConflict(nextSpender); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If this tx spends any previous credits, set each unspent.
|
|
|
|
for _, input := range r.Tx().MsgTx().TxIn {
|
|
|
|
delete(u.previousOutpoints, input.PreviousOutpoint)
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// For all mined transactions with credits spent by this
|
|
|
|
// conflicting transaction, remove from the bookkeeping maps
|
|
|
|
// and set each previous record's credit as unspent.
|
|
|
|
prevKey, ok := u.spentBlockOutPointKeys[input.PreviousOutpoint]
|
|
|
|
if ok {
|
|
|
|
delete(u.spentBlockOutPointKeys, input.PreviousOutpoint)
|
|
|
|
delete(u.spentBlockOutPoints, prevKey)
|
|
|
|
prev, err := s.lookupBlockTx(prevKey.BlockTxKey)
|
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
prev.credits[prevKey.OutputIndex].spentBy = nil
|
|
|
|
continue
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// For all unmined transactions with credits spent by this
|
|
|
|
// conflicting transaction, remove from the unspent store's
|
|
|
|
// spent tracking.
|
|
|
|
//
|
|
|
|
// Spend tracking is only handled by these maps, so there is
|
|
|
|
// no need to modify the record and unset a spent-by pointer.
|
|
|
|
if _, ok := u.spentUnconfirmed[input.PreviousOutpoint]; ok {
|
|
|
|
delete(u.spentUnconfirmed, input.PreviousOutpoint)
|
|
|
|
}
|
|
|
|
}
|
2013-09-09 20:14:57 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
delete(u.txs, *r.Tx().Sha())
|
|
|
|
for _, input := range r.Tx().MsgTx().TxIn {
|
|
|
|
delete(u.previousOutpoints, input.PreviousOutpoint)
|
|
|
|
}
|
2014-06-20 01:11:47 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return nil
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// UnspentOutputs returns all unspent received transaction outputs.
|
|
|
|
// The order is undefined.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (s *Store) UnspentOutputs() ([]Credit, error) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
s.mtx.RLock()
|
|
|
|
defer s.mtx.RUnlock()
|
|
|
|
|
|
|
|
return s.unspentOutputs()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (s *Store) unspentOutputs() ([]Credit, error) {
|
2014-06-18 05:48:54 +02:00
|
|
|
type createdCredit struct {
|
2014-06-18 07:16:08 +02:00
|
|
|
credit Credit
|
2014-06-18 05:48:54 +02:00
|
|
|
err error
|
|
|
|
}
|
|
|
|
|
|
|
|
creditChans := make([]chan createdCredit, len(s.unspent))
|
|
|
|
i := 0
|
|
|
|
for op, key := range s.unspent {
|
|
|
|
creditChans[i] = make(chan createdCredit)
|
2014-06-18 23:46:18 +02:00
|
|
|
go func(i int, key BlockTxKey, opIndex uint32) {
|
2014-06-18 05:48:54 +02:00
|
|
|
r, err := s.lookupBlockTx(key)
|
|
|
|
if err != nil {
|
|
|
|
creditChans[i] <- createdCredit{err: err}
|
|
|
|
return
|
|
|
|
}
|
|
|
|
t := &TxRecord{key, r, s}
|
2014-06-18 07:16:08 +02:00
|
|
|
c := Credit{t, opIndex}
|
2014-06-18 05:48:54 +02:00
|
|
|
creditChans[i] <- createdCredit{credit: c}
|
2014-06-18 23:46:18 +02:00
|
|
|
}(i, key, op.Index)
|
2014-06-18 05:48:54 +02:00
|
|
|
i++
|
|
|
|
}
|
|
|
|
|
2014-06-18 16:16:14 +02:00
|
|
|
unspent := make([]Credit, len(s.unspent))
|
|
|
|
for i, c := range creditChans {
|
|
|
|
cc := <-c
|
2014-06-18 05:48:54 +02:00
|
|
|
if cc.err != nil {
|
|
|
|
return nil, cc.err
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2014-06-18 16:16:14 +02:00
|
|
|
unspent[i] = cc.credit
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-06-18 05:48:54 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for _, r := range s.unconfirmed.txs {
|
|
|
|
for outputIndex, credit := range r.credits {
|
|
|
|
if credit == nil || credit.spentBy != nil {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
key := BlockTxKey{BlockHeight: -1}
|
|
|
|
txRecord := &TxRecord{key, r, s}
|
2014-06-18 07:16:08 +02:00
|
|
|
c := Credit{txRecord, uint32(outputIndex)}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
unspent = append(unspent, c)
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
}
|
2014-06-18 05:48:54 +02:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return unspent, nil
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
2014-05-23 17:31:05 +02:00
|
|
|
// unspentTx is a type defined here so it can be sorted with the sort
|
|
|
|
// package. It is used to provide a sorted range over all transactions
|
|
|
|
// in a block with unspent outputs.
|
|
|
|
type unspentTx struct {
|
|
|
|
blockIndex int
|
|
|
|
sliceIndex uint32
|
|
|
|
}
|
|
|
|
|
2014-06-18 07:16:08 +02:00
|
|
|
type creditSlice []Credit
|
2014-05-23 17:31:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
func (s creditSlice) Len() int {
|
|
|
|
return len(s)
|
|
|
|
}
|
2014-05-23 17:31:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
func (s creditSlice) Less(i, j int) bool {
|
|
|
|
switch {
|
|
|
|
// If both credits are from the same tx, sort by output index.
|
|
|
|
case s[i].Tx().Sha() == s[j].Tx().Sha():
|
|
|
|
return s[i].OutputIndex < s[j].OutputIndex
|
2014-05-23 17:31:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
// If both transactions are unmined, sort by their received date.
|
|
|
|
case s[i].BlockIndex == -1 && s[j].BlockIndex == -1:
|
|
|
|
return s[i].received.Before(s[j].received)
|
2014-05-23 17:31:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
// Unmined (newer) txs always come last.
|
|
|
|
case s[i].BlockIndex == -1:
|
|
|
|
return false
|
|
|
|
case s[j].BlockIndex == -1:
|
|
|
|
return true
|
2014-05-23 17:31:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
// If both txs are mined in different blocks, sort by block height.
|
|
|
|
case s[i].BlockHeight != s[j].BlockHeight:
|
|
|
|
return s[i].BlockHeight < s[j].BlockHeight
|
2014-05-23 17:31:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
// Both txs share the same block, sort by block index.
|
|
|
|
default:
|
|
|
|
return s[i].BlockIndex < s[j].BlockIndex
|
|
|
|
}
|
|
|
|
}
|
2014-05-23 17:31:05 +02:00
|
|
|
|
2014-06-18 05:48:54 +02:00
|
|
|
func (s creditSlice) Swap(i, j int) {
|
|
|
|
s[i], s[j] = s[j], s[i]
|
|
|
|
}
|
2014-05-23 17:31:05 +02:00
|
|
|
|
|
|
|
// SortedUnspentOutputs returns all unspent recevied transaction outputs.
|
|
|
|
// The order is first unmined transactions (sorted by receive date), then
|
|
|
|
// mined transactions in increasing number of confirmations. Transactions
|
|
|
|
// in the same block (same number of confirmations) are sorted by block
|
|
|
|
// index in increasing order. Credits (outputs) from the same transaction
|
|
|
|
// are sorted by output index in increasing order.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (s *Store) SortedUnspentOutputs() ([]Credit, error) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
s.mtx.RLock()
|
|
|
|
defer s.mtx.RUnlock()
|
|
|
|
|
|
|
|
unspent, err := s.unspentOutputs()
|
2014-06-18 05:48:54 +02:00
|
|
|
if err != nil {
|
2014-06-18 07:16:08 +02:00
|
|
|
return []Credit{}, err
|
2014-05-23 17:31:05 +02:00
|
|
|
}
|
2014-06-18 05:48:54 +02:00
|
|
|
sort.Sort(sort.Reverse(creditSlice(unspent)))
|
2014-05-23 17:31:05 +02:00
|
|
|
return unspent, nil
|
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// confirmed checks whether a transaction at height txHeight has met
|
|
|
|
// minconf confirmations for a blockchain at height curHeight.
|
|
|
|
func confirmed(minconf int, txHeight, curHeight int32) bool {
|
|
|
|
return confirms(txHeight, curHeight) >= int32(minconf)
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// confirms returns the number of confirmations for a transaction in a
|
|
|
|
// block at height txHeight (or -1 for an unconfirmed tx) given the chain
|
|
|
|
// height curHeight.
|
|
|
|
func confirms(txHeight, curHeight int32) int32 {
|
|
|
|
switch {
|
|
|
|
case txHeight == -1, txHeight > curHeight:
|
|
|
|
return 0
|
|
|
|
default:
|
|
|
|
return curHeight - txHeight + 1
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Balance returns the spendable wallet balance (total value of all unspent
|
|
|
|
// transaction outputs) given a minimum of minConf confirmations, calculated
|
|
|
|
// at a current chain height of curHeight. Coinbase outputs are only included
|
|
|
|
// in the balance if maturity has been reached.
|
|
|
|
func (s *Store) Balance(minConf int, chainHeight int32) (btcutil.Amount, error) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
s.mtx.RLock()
|
|
|
|
defer s.mtx.RUnlock()
|
|
|
|
|
|
|
|
return s.balance(minConf, chainHeight)
|
|
|
|
}
|
|
|
|
|
|
|
|
func (s *Store) balance(minConf int, chainHeight int32) (btcutil.Amount, error) {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
var bal btcutil.Amount
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Shadow these functions to avoid repeating arguments unnecesarily.
|
|
|
|
confirms := func(height int32) int32 {
|
|
|
|
return confirms(height, chainHeight)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
confirmed := func(height int32) bool {
|
|
|
|
return confirmed(minConf, height, chainHeight)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
|
|
|
|
for _, b := range s.blocks {
|
|
|
|
if confirmed(b.Height) {
|
|
|
|
bal += b.amountDeltas.Spendable
|
|
|
|
if confirms(b.Height) >= btcchain.CoinbaseMaturity {
|
|
|
|
bal += b.amountDeltas.Reward
|
|
|
|
}
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
// If there are still blocks that contain debiting transactions,
|
|
|
|
// decrement the balance if they spend credits meeting minConf
|
|
|
|
// confirmations.
|
|
|
|
for _, r := range b.txs {
|
|
|
|
if r.debits == nil {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
for _, prev := range r.debits.spends {
|
|
|
|
if !confirmed(prev.BlockHeight) {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
r, err := s.lookupBlockTx(prev.BlockTxKey)
|
|
|
|
if err != nil {
|
|
|
|
return 0, err
|
|
|
|
}
|
|
|
|
v := r.Tx().MsgTx().TxOut[prev.OutputIndex].Value
|
|
|
|
bal -= btcutil.Amount(v)
|
|
|
|
}
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Unconfirmed transactions which spend previous credits debit from
|
|
|
|
// the returned balance, even with minConf > 0.
|
|
|
|
for prev := range s.unconfirmed.spentBlockOutPoints {
|
|
|
|
if confirmed(prev.BlockHeight) {
|
|
|
|
r, err := s.lookupBlockTx(prev.BlockTxKey)
|
|
|
|
if err != nil {
|
|
|
|
return 0, err
|
|
|
|
}
|
|
|
|
v := r.Tx().MsgTx().TxOut[prev.OutputIndex].Value
|
|
|
|
bal -= btcutil.Amount(v)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// If unconfirmed credits are included, tally them as well.
|
|
|
|
if minConf == 0 {
|
|
|
|
for _, r := range s.unconfirmed.txs {
|
|
|
|
for i, c := range r.credits {
|
|
|
|
if c == nil {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
if c.spentBy == nil {
|
|
|
|
v := r.Tx().MsgTx().TxOut[i].Value
|
|
|
|
bal += btcutil.Amount(v)
|
|
|
|
}
|
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return bal, nil
|
2014-01-27 18:53:32 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Records returns a chronologically-ordered slice of all transaction records
|
|
|
|
// saved by the store. This is sorted first by block height in increasing
|
|
|
|
// order, and then by transaction index for each tx in a block.
|
|
|
|
func (s *Store) Records() (records []*TxRecord) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
s.mtx.RLock()
|
|
|
|
defer s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for _, b := range s.blocks {
|
|
|
|
for _, r := range b.txs {
|
|
|
|
key := BlockTxKey{r.tx.Index(), b.Block.Height}
|
|
|
|
records = append(records, &TxRecord{key, r, s})
|
|
|
|
}
|
|
|
|
}
|
2014-02-03 19:29:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Unconfirmed records are saved unsorted, and must be sorted by
|
|
|
|
// received date on the fly.
|
2014-06-13 23:17:15 +02:00
|
|
|
unconfirmed := make([]*TxRecord, 0, len(s.unconfirmed.txs))
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for _, r := range s.unconfirmed.txs {
|
|
|
|
key := BlockTxKey{BlockHeight: -1}
|
2014-06-13 23:17:15 +02:00
|
|
|
unconfirmed = append(unconfirmed, &TxRecord{key, r, s})
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
|
|
|
sort.Sort(byReceiveDate(unconfirmed))
|
|
|
|
records = append(records, unconfirmed...)
|
2014-02-03 19:29:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return
|
2014-02-03 19:29:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Implementation of sort.Interface to sort transaction records by their
|
|
|
|
// receive date.
|
|
|
|
type byReceiveDate []*TxRecord
|
2014-02-03 19:29:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
func (r byReceiveDate) Len() int { return len(r) }
|
|
|
|
func (r byReceiveDate) Less(i, j int) bool { return r[i].received.Before(r[j].received) }
|
|
|
|
func (r byReceiveDate) Swap(i, j int) { r[i], r[j] = r[j], r[i] }
|
2014-02-03 19:29:25 +01:00
|
|
|
|
2014-06-18 07:16:08 +02:00
|
|
|
// Debits returns the debit record for the transaction, or a non-nil error if
|
|
|
|
// the transaction does not debit from any previous transaction credits.
|
|
|
|
func (t *TxRecord) Debits() (Debits, error) {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
if t.debits == nil {
|
2014-06-18 07:16:08 +02:00
|
|
|
return Debits{}, errors.New("no debits")
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
2014-06-18 07:16:08 +02:00
|
|
|
return Debits{t}, nil
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Credits returns all credit records for this transaction's outputs that are or
|
|
|
|
// were spendable by wallet.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (t *TxRecord) Credits() []Credit {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
2014-06-18 07:16:08 +02:00
|
|
|
credits := make([]Credit, 0, len(t.credits))
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
for i, c := range t.credits {
|
|
|
|
if c != nil {
|
2014-06-18 07:16:08 +02:00
|
|
|
credits = append(credits, Credit{t, uint32(i)})
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return credits
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
|
|
|
|
2014-06-18 23:46:18 +02:00
|
|
|
// HasCredit returns whether the transaction output at the passed index is
|
|
|
|
// a wallet credit.
|
|
|
|
func (t *TxRecord) HasCredit(i int) bool {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
2014-06-18 23:46:18 +02:00
|
|
|
if len(t.credits) <= i {
|
|
|
|
return false
|
|
|
|
}
|
|
|
|
return t.credits[i] != nil
|
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// InputAmount returns the total amount debited from previous credits.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (d Debits) InputAmount() btcutil.Amount {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
d.s.mtx.RLock()
|
|
|
|
defer d.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return d.txRecord.debits.amount
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// OutputAmount returns the total amount of all outputs for a transaction.
|
|
|
|
func (t *TxRecord) OutputAmount(ignoreChange bool) btcutil.Amount {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
|
|
|
return t.outputAmount(ignoreChange)
|
|
|
|
}
|
|
|
|
|
|
|
|
func (t *TxRecord) outputAmount(ignoreChange bool) btcutil.Amount {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
a := btcutil.Amount(0)
|
|
|
|
for i, txOut := range t.Tx().MsgTx().TxOut {
|
2014-05-08 05:48:18 +02:00
|
|
|
if ignoreChange {
|
|
|
|
switch cs := t.credits; {
|
|
|
|
case i < len(cs) && cs[i] != nil && cs[i].change:
|
|
|
|
continue
|
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-05-08 05:48:18 +02:00
|
|
|
a += btcutil.Amount(txOut.Value)
|
2013-11-22 19:42:25 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return a
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Fee returns the difference between the debited amount and the total
|
|
|
|
// transaction output.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (d Debits) Fee() btcutil.Amount {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
return d.txRecord.debits.amount - d.outputAmount(false)
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
2013-11-22 19:42:25 +01:00
|
|
|
|
2014-02-24 20:35:30 +01:00
|
|
|
// Addresses parses the pubkey script, extracting all addresses for a
|
|
|
|
// standard script.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (c Credit) Addresses(net *btcnet.Params) (btcscript.ScriptClass,
|
2014-02-24 20:35:30 +01:00
|
|
|
[]btcutil.Address, int, error) {
|
2013-11-22 19:42:25 +01:00
|
|
|
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
c.s.mtx.RLock()
|
|
|
|
defer c.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
msgTx := c.Tx().MsgTx()
|
|
|
|
pkScript := msgTx.TxOut[c.OutputIndex].PkScript
|
2014-05-27 19:50:51 +02:00
|
|
|
return btcscript.ExtractPkScriptAddrs(pkScript, net)
|
2013-08-22 19:14:21 +02:00
|
|
|
}
|
2014-01-27 18:53:32 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Change returns whether the credit is the result of a change output.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (c Credit) Change() bool {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
c.s.mtx.RLock()
|
|
|
|
defer c.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return c.txRecord.credits[c.OutputIndex].change
|
2014-01-27 18:53:32 +01:00
|
|
|
}
|
2014-02-03 19:29:25 +01:00
|
|
|
|
2014-05-07 05:48:12 +02:00
|
|
|
// Confirmed returns whether a transaction has reached some target number of
|
|
|
|
// confirmations, given the current best chain height.
|
|
|
|
func (t *TxRecord) Confirmed(target int, chainHeight int32) bool {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
2014-05-07 05:48:12 +02:00
|
|
|
return confirmed(target, t.BlockHeight, chainHeight)
|
|
|
|
}
|
|
|
|
|
|
|
|
// Confirmations returns the total number of confirmations a transaction has
|
|
|
|
// reached, given the current best chain height.
|
|
|
|
func (t *TxRecord) Confirmations(chainHeight int32) int32 {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
2014-05-07 05:48:12 +02:00
|
|
|
return confirms(t.BlockHeight, chainHeight)
|
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// IsCoinbase returns whether the transaction is a coinbase.
|
|
|
|
func (t *TxRecord) IsCoinbase() bool {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
t.s.mtx.RLock()
|
|
|
|
defer t.s.mtx.RUnlock()
|
|
|
|
|
|
|
|
return t.isCoinbase()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (t *TxRecord) isCoinbase() bool {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return t.BlockHeight != -1 && t.BlockIndex == 0
|
2014-02-03 19:29:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Amount returns the amount credited to the account from a transaction output.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (c Credit) Amount() btcutil.Amount {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
c.s.mtx.RLock()
|
|
|
|
defer c.s.mtx.RUnlock()
|
|
|
|
|
|
|
|
return c.amount()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (c Credit) amount() btcutil.Amount {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
msgTx := c.Tx().MsgTx()
|
|
|
|
return btcutil.Amount(msgTx.TxOut[c.OutputIndex].Value)
|
2014-02-03 19:29:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// OutPoint returns the outpoint needed to include in a transaction input
|
|
|
|
// to spend this output.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (c Credit) OutPoint() *btcwire.OutPoint {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
c.s.mtx.RLock()
|
|
|
|
defer c.s.mtx.RUnlock()
|
|
|
|
|
|
|
|
return c.outPoint()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (c Credit) outPoint() *btcwire.OutPoint {
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return btcwire.NewOutPoint(c.Tx().Sha(), c.OutputIndex)
|
2014-02-03 19:29:25 +01:00
|
|
|
}
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// outputKey creates and returns the block lookup key for this credit.
|
2014-06-18 23:46:18 +02:00
|
|
|
func (c Credit) outputKey() BlockOutputKey {
|
|
|
|
return BlockOutputKey{
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
BlockTxKey: c.BlockTxKey,
|
|
|
|
OutputIndex: c.OutputIndex,
|
2014-02-24 20:35:30 +01:00
|
|
|
}
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
}
|
2014-02-24 20:35:30 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Spent returns whether the transaction output is currently spent or not.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (c Credit) Spent() bool {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
c.s.mtx.RLock()
|
|
|
|
defer c.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return c.txRecord.credits[c.OutputIndex].spentBy != nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// TxOut returns the transaction output which this credit references.
|
2014-06-18 07:16:08 +02:00
|
|
|
func (c Credit) TxOut() *btcwire.TxOut {
|
Remove account support, fix races on btcd connect.
This commit is the result of several big changes being made to the
wallet. In particular, the "handshake" (initial sync to the chain
server) was quite racy and required proper synchronization. To make
fixing this race easier, several other changes were made to the
internal wallet data structures and much of the RPC server ended up
being rewritten.
First, all account support has been removed. The previous Account
struct has been replaced with a Wallet structure, which includes a
keystore for saving keys, and a txstore for storing relevant
transactions. This decision has been made since it is the opinion of
myself and other developers that bitcoind accounts are fundamentally
broken (as accounts implemented by bitcoind support both arbitrary
address groupings as well as moving balances between accounts -- these
are fundamentally incompatible features), and since a BIP0032 keystore
is soon planned to be implemented (at which point, "accounts" can
return as HD extended keys). With the keystore handling the grouping
of related keys, there is no reason have many different Account
structs, and the AccountManager has been removed as well. All RPC
handlers that take an account option will only work with "" (the
default account) or "*" if the RPC allows specifying all accounts.
Second, much of the RPC server has been cleaned up. The global
variables for the RPC server and chain server client have been moved
to part of the rpcServer struct, and the handlers for each RPC method
that are looked up change depending on which components have been set.
Passthrough requests are also no longer handled specially, but when
the chain server is set, a handler to perform the passthrough will be
returned if the method is not otherwise a wallet RPC. The
notification system for websocket clients has also been rewritten so
wallet components can send notifications through channels, rather than
requiring direct access to the RPC server itself, or worse still,
sending directly to a websocket client's send channel. In the future,
this will enable proper registration of notifications, rather than
unsolicited broadcasts to every connected websocket client (see
issue #84).
Finally, and the main reason why much of this cleanup was necessary,
the races during intial sync with the chain server have been fixed.
Previously, when the 'Handshake' was run, a rescan would occur which
would perform modifications to Account data structures as
notifications were received. Synchronization was provided with a
single binary semaphore which serialized all access to wallet and
account data. However, the Handshake itself was not able to run with
this lock (or else notifications would block), and many data races
would occur as both notifications were being handled. If GOMAXPROCS
was ever increased beyond 1, btcwallet would always immediately crash
due to invalid addresses caused by the data races on startup. To fix
this, the single lock for all wallet access has been replaced with
mutexes for both the keystore and txstore. Handling of btcd
notifications and client requests may now occur simultaneously.
GOMAXPROCS has also been set to the number of logical CPUs at the
beginning of main, since with the data races fixed, there's no reason
to prevent the extra parallelism gained by increasing it.
Closes #78.
Closes #101.
Closes #110.
2014-07-09 05:17:38 +02:00
|
|
|
c.s.mtx.RLock()
|
|
|
|
defer c.s.mtx.RUnlock()
|
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
return c.Tx().MsgTx().TxOut[c.OutputIndex]
|
|
|
|
}
|
2014-02-03 19:29:25 +01:00
|
|
|
|
Another day, another tx store implementation.
The last transaction store was a great example of how not to write
scalable software. For a variety of reasons, it was very slow at
processing transaction inserts. Among them:
1) Every single transaction record being saved in a linked list
(container/list), and inserting into this list would be an O(n)
operation so that records could be ordered by receive date.
2) Every single transaction in the above mentioned list was iterated
over in order to find double spends which must be removed. It is
silly to do this check for mined transactions, which already have
been checked for this by btcd. Worse yet, if double spends were
found, the list would be iterated a second (or third, or fourth)
time for each removed transaction.
3) All spend tracking for signed-by-wallet transactions was found on
each transaction insert, even if the now spent previous transaction
outputs were known by the caller.
This list could keep going on, but you get the idea. It was bad.
To resolve these issues a new transaction store had to be implemented.
The new implementation:
1) Tracks mined and unmined transactions in different data structures.
Mined transactions are cheap to track because the required double
spend checks have already been performed by the chain server, and
double spend checks are only required to be performed on
newly-inserted mined transactions which may conflict with previous
unmined transactions.
2) Saves mined transactions grouped by block first, and then by their
transaction index. Lookup keys for mined transactions are simply
the block height (in the best chain, that's all we save) and index
of the transaction in the block. This makes looking up any
arbitrary transaction almost an O(1) operation (almost, because
block height and block indexes are mapped to their slice indexes
with a Go map).
3) Saves records in each transaction for whether the outputs are
wallet credits (spendable by wallet) and for whether inputs debit
from previous credits. Both structures point back to the source
or spender (credits point to the transaction that spends them, or
nil for unspent credits, and debits include keys to lookup the
transaction credits they spent. While complicated to keep track
of, this greatly simplifies the spent tracking for transactions
across rollbacks and transaction removals.
4) Implements double spend checking as an almost O(1) operation. A
Go map is used to map each previous outpoint for all unconfirmed
transactions to the unconfirmed tx record itself. Checking for
double spends on confirmed transaction inserts only involves
looking up each previous outpoint of the inserted tx in this map.
If a double spend is found, removal is simplified by only
removing the transaction and its spend chain from store maps,
rather than iterating a linked list several times over to remove
each dead transaction in the spend chain.
5) Allows the caller to specify the previous credits which are spent
by a debiting transaction. When a transaction is created by
wallet, the previous outputs are already known, and by passing
their record types to the AddDebits method, lookups for each
previously unspent credit are omitted.
6) Bookkeeps all blocks with transactions with unspent credits, and
bookkeeps the transaction indexes of all transactions with unspent
outputs for a single block. For the case where the caller adding a
debit record does not know what credits a transaction debits from,
these bookkeeping structures allow the store to only consider known
unspent transactions, rather than searching through both spent and
unspents.
7) Saves amount deltas for the entire balance as a result of each
block, due to transactions within that block. This improves the
performance of calculating the full balance by not needing to
iterate over every transaction, and then every credit, to determine
if a credit is spent or unspent. When transactions are moved from
unconfirmed to a block structure, the amount deltas are incremented
by the amount of all transaction credits (both spent and unspent)
and debited by the total amount the transaction spends from
previous wallet credits. For the common case of calculating a
balance with just one confirmation, the only involves iterating
over each block structure and adding the (possibly negative)
amount delta. Coinbase rewards are saved similarly, but with a
different amount variable so they can be seperatly included or
excluded.
Due to all of the changes in how the store internally works, the
serialization format has changed. To simplify the serialization
logic, support for reading the last store file version has been
removed. Past this change, a rescan (run automatically) will be
required to rebuild the transaction history.
2014-05-05 23:12:05 +02:00
|
|
|
// Tx returns the underlying transaction.
|
|
|
|
func (r *txRecord) Tx() *btcutil.Tx {
|
|
|
|
return r.tx
|
2014-02-03 19:29:25 +01:00
|
|
|
}
|