dd435ad Add unit tests for signals generated by ProcessNewBlock() (Jesse Cohen)
a3ae8e6 Fix concurrency-related bugs in ActivateBestChain (Jesse Cohen)
ecc3c4a Do not unlock cs_main in ABC unless we've actually made progress. (Matt Corallo)
Pull request description:
Originally this PR was just to add tests around concurrency in block validation - those tests seem to have uncovered another bug in ActivateBestChain - this now fixes that bug and adds tests.
ActivateBestChain (invoked after a new block is validated) proceeds in steps - acquiring and releasing cs_main while incrementally disconnecting and connecting blocks to sync to the most work chain known (FindMostWorkChain()). Every time cs_main is released the result of FindMostWorkChain() can change - but currently that value is cached across acquisitions of cs_main and only refreshed when an invalid chain is explored. It needs to be refreshed every time cs_main is reacquired. The test added in 6094ce7304 will occasionally fail without the commit fixing this issue 26bfdbaddb
Original description below
--
After a bug discovered where UpdatedBlockTip() notifications could be triggered out of order (#12978), these unit tests check certain invariants about these signals.
The scheduler test asserts that a SingleThreadedSchedulerClient processes callbacks fully and sequentially.
The block validation test generates a random chain and calls ProcessNewBlock from multiple threads at random and in parallel. ValidationInterface callbacks verify that the ordering of BlockConnected BlockDisconnected and UpdatedBlockTip events occur as expected.
Tree-SHA512: 4102423a03d2ea28580c7a70add8a6bdb22ef9e33b107c3aadef80d5af02644cdfaae516c44933924717599c81701e0b96fbf9cf38696e9e41372401a5ee1f3c
Technically, some internal datastructures may be in an inconsistent
state if we do this, though there are no known bugs there. Still,
for future safety, its much better to only unlock cs_main if we've
made progress (not just tried a reorg which may make progress).
Previously, ChainStateFlushed would fire either if a full flush
completed (which can happen due to memory limits, forced flush, or
on its own DATABASE_WRITE_INTERVAL timer) *or* on a
ChainStateFlushed-specific DATABASE_WRITE_INTERVAL timer. This is
both less clear for clients (as there are no guarantees about a
flush having actually happened prior to the call), and reults in
extra flushes not clearly intended by the code. We drop the second
case, providing a strong guarantee without removing the periodit
timer-based flushing.
This much more accurately captures the meaning of the callback.
-BEGIN VERIFY SCRIPT-
sed -i 's/SetBestChain/ChainStateFlushed/g' src/validationinterface.h src/validationinterface.cpp src/wallet/wallet.h src/wallet/wallet.cpp src/validation.cpp src/index/txindex.h src/index/txindex.cpp
-END VERIFY SCRIPT-
* Make PeerLogicValidation final to prevent deriving from it [1]
* Prevent deletions of NetEventsInterface and CValidationInterface
objects via a base class pointer
[1] silences the following compiler warning (from Clang 7.0.0):
/usr/include/c++/v1/memory:2285:5: error: delete called on non-final 'PeerLogicValidation' that has
virtual functions but non-virtual destructor [-Werror,-Wdelete-non-virtual-dtor]
delete __ptr;
^
/usr/include/c++/v1/memory:2598:7: note: in instantiation of member function
'std::__1::default_delete<PeerLogicValidation>::operator()' requested here
__ptr_.second()(__tmp);
^
init.cpp:201:15: note: in instantiation of member function 'std::__1::unique_ptr<PeerLogicValidation,
std::__1::default_delete<PeerLogicValidation> >::reset' requested here
peerLogic.reset();
^
Note that UpdatedBlockTip is also used in net_processing to
announce new blocks to peers. As this may need additional review,
this change is included in its own commit.
This runs Block{Connected,Disconnected}, SetBestChain, Inventory,
and TransactionAddedToMempool on the background scheduler thread.
Of those, only BlockConnected is used outside of Wallet/ZMQ, and
is used only for orphan transaction removal in net_processing,
something which does not need to be synchronous with anything
else.
This partially reverts #9583, re-enabling some of the gains from
#7946. This does not, however, re-enable the gains achieved by
repeatedly releasing cs_main between each transaction processed.
This is both good practice (we want to move all such callbacks
into a background thread eventually) and prevents a lock inversion
when we go to use this in wallet (mempool.cs->cs_wallet and
cs_wallet->mempool.cs would otherwise both be used).
This is currently unused, but will by used by wallet to cache when
transactions are in the mempool, obviating the need for calls to
mempool from CWalletTx::InMempool()
Note that the CScheduler thread cant be running at this point,
it has already been stopped with the rest of the init threadgroup.
Thus, just calling any remaining loose callbacks during Shutdown()
is sane.
This will be used by CValidationInterface soon.
This requires a bit of work as we need to ensure that most of our
callbacks happen in-order (to avoid synchronization issues in
wallet) - we keep our own internal queue and push things onto it,
scheduling a queue-draining function immediately upon new
callbacks.
(by hiding boost::signals stuff in the .cpp)
This allows us to give it a bit more intelligence as we move
forward, including routing some signals through CScheduler. While
the introduction of a "internals" pointer in the class is pretty
ugly, the fact that we no longer need to include boost/signals
directly from validationinterface.h is very much worth the loss.
This makes it possible to mine to any wallet when multi-wallet mode is added.
Solves the same problem as #10649, but IMO in a cleaner way.
It also gets rid of the circuitous `ScriptForMining` method on
`CValidationInterface`, which really doesn't belong there.
After this change it's still possible to mine without wallet through
`generatetoaddress`.
This removes another callback from block connection logic, making it
easier to reason about the wallet-RPCs-returns-stale-info issue.
UpdatedTransaction was previously used by the GUI to display
coinbase transactions only after they have a block built on top of
them. This worked fine for in most cases, but only worked due to a
corner case if the user received a coinbase payout in a block
immediately prior to restart. In that case, the normal process of
caching the most recent coinbase transaction's hash would not work,
and instead it would only work because of the on-load -checkblocks
calling DisconnectBlock and ConnectBlock on the current tip.
In order to make this more robust, a full mapWallet loop after the
first block which is connected after restart was added.
This simplifies fixing the wallet-returns-stale-info issue as we
can now hold cs_wallet across an entire block instead of only
per-tx (though we only actually do so in the next commit).
This change also removes the NOT_IN_BLOCK constant in favor of only
passing the CBlockIndex* parameter to SyncTransactions when a new
block is being connected, instead of also when a block is being
disconnected.
This change adds a parameter to BlockConnectedDisconnected which
lists the transactions which were removed from mempool due to
confliction as a result of this operation. While its somewhat of a
shame to make block-validation-logic generate a list of mempool
changes to be included in its generated callbacks, fixing this isnt
too hard.
Further in this change-set, CValidationInterface starts listening
to mempool directly, placing it in the middle and giving it a bit
of logic to know how to route notifications from block-validation,
mempool, etc (though not listening for conflicted-removals yet).
c4a6929 Clarify assumptions made about when BlockCheck is called (Matt Corallo)
Tree-SHA512: 2eceb0c4f06c7fd6b290b93843bda11a4b63131559c5e8226bfec84596ed4e54ee6d8f5bc9cf789a80675be8b8079cf9234c96032df306258cb2260b9d8c7825
In anticipation of making all the callbacks out of block processing
flow through it. Note that vHashes will always have something in it
since pindexFork != pindexNewTip.
- removes mapBlockIndex find operation
- theoretically allows removing the cs_main lock during zqm notification while introducing a new file position lock
Adds a regression test for the wallet's ResendWalletTransactions function, which uses a new, hidden RPC command "resendwallettransactions."
I refactored main's Broadcast signal so it is passed the best-block time, which let me remove a global variable shared between main.cpp and the wallet (nTimeBestReceived).
I also manually tested the "rebroadcast unconfirmed every half hour or so" functionality by:
1. Running bitcoind -connect=0.0.0.0:8333
2. Creating a couple of send-to-self transactions
3. Connect to a peer using -addnode
4. Waited a while, monitoring debug.log, until I see:
```2015-03-23 18:48:10 ResendWalletTransactions: rebroadcast 2 unconfirmed transactions```
One last change: don't bother putting ResendWalletTransactions messages in debug.log unless unconfirmed transactions were actually rebroadcast.
