lbcd/database/ffldb/whitebox_test.go

706 lines
19 KiB
Go
Raw Normal View History

database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
// Copyright (c) 2015-2016 The btcsuite developers
// Use of this source code is governed by an ISC
// license that can be found in the LICENSE file.
// This file is part of the ffldb package rather than the ffldb_test package as
// it provides whitebox testing.
package ffldb
import (
"compress/bzip2"
"encoding/binary"
"fmt"
"hash/crc32"
"io"
"os"
"path/filepath"
"testing"
"github.com/btcsuite/goleveldb/leveldb"
ldberrors "github.com/btcsuite/goleveldb/leveldb/errors"
"github.com/lbryio/lbcd/database"
"github.com/lbryio/lbcd/wire"
btcutil "github.com/lbryio/lbcutil"
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
)
var (
// blockDataNet is the expected network in the test block data.
blockDataNet = wire.MainNet
// blockDataFile is the path to a file containing the first 256 blocks
// of the block chain.
blockDataFile = filepath.Join("..", "testdata", "blocks1-256.bz2")
// errSubTestFail is used to signal that a sub test returned false.
errSubTestFail = fmt.Errorf("sub test failure")
)
// loadBlocks loads the blocks contained in the testdata directory and returns
// a slice of them.
func loadBlocks(t *testing.T, dataFile string, network wire.BitcoinNet) ([]*btcutil.Block, error) {
// Open the file that contains the blocks for reading.
fi, err := os.Open(dataFile)
if err != nil {
t.Errorf("failed to open file %v, err %v", dataFile, err)
return nil, err
}
defer func() {
if err := fi.Close(); err != nil {
t.Errorf("failed to close file %v %v", dataFile,
err)
}
}()
dr := bzip2.NewReader(fi)
// Set the first block as the genesis block.
blocks := make([]*btcutil.Block, 0, 256)
genesis := btcutil.NewBlock(chaincfg.MainNetParams.GenesisBlock)
blocks = append(blocks, genesis)
// Load the remaining blocks.
for height := 1; ; height++ {
var net uint32
err := binary.Read(dr, binary.LittleEndian, &net)
if err == io.EOF {
// Hit end of file at the expected offset. No error.
break
}
if err != nil {
t.Errorf("Failed to load network type for block %d: %v",
height, err)
return nil, err
}
if net != uint32(network) {
t.Errorf("Block doesn't match network: %v expects %v",
net, network)
return nil, err
}
var blockLen uint32
err = binary.Read(dr, binary.LittleEndian, &blockLen)
if err != nil {
t.Errorf("Failed to load block size for block %d: %v",
height, err)
return nil, err
}
// Read the block.
blockBytes := make([]byte, blockLen)
_, err = io.ReadFull(dr, blockBytes)
if err != nil {
t.Errorf("Failed to load block %d: %v", height, err)
return nil, err
}
// Deserialize and store the block.
block, err := btcutil.NewBlockFromBytes(blockBytes)
if err != nil {
t.Errorf("Failed to parse block %v: %v", height, err)
return nil, err
}
blocks = append(blocks, block)
}
return blocks, nil
}
// checkDbError ensures the passed error is a database.Error with an error code
// that matches the passed error code.
func checkDbError(t *testing.T, testName string, gotErr error, wantErrCode database.ErrorCode) bool {
dbErr, ok := gotErr.(database.Error)
if !ok {
t.Errorf("%s: unexpected error type - got %T, want %T",
testName, gotErr, database.Error{})
return false
}
if dbErr.ErrorCode != wantErrCode {
t.Errorf("%s: unexpected error code - got %s (%s), want %s",
testName, dbErr.ErrorCode, dbErr.Description,
wantErrCode)
return false
}
return true
}
// testContext is used to store context information about a running test which
// is passed into helper functions.
type testContext struct {
t *testing.T
db database.DB
files map[uint32]*lockableFile
maxFileSizes map[uint32]int64
blocks []*btcutil.Block
}
// TestConvertErr ensures the leveldb error to database error conversion works
// as expected.
func TestConvertErr(t *testing.T) {
t.Parallel()
tests := []struct {
err error
wantErrCode database.ErrorCode
}{
{&ldberrors.ErrCorrupted{}, database.ErrCorruption},
{leveldb.ErrClosed, database.ErrDbNotOpen},
{leveldb.ErrSnapshotReleased, database.ErrTxClosed},
{leveldb.ErrIterReleased, database.ErrTxClosed},
}
for i, test := range tests {
gotErr := convertErr("test", test.err)
if gotErr.ErrorCode != test.wantErrCode {
t.Errorf("convertErr #%d unexpected error - got %v, "+
"want %v", i, gotErr.ErrorCode, test.wantErrCode)
continue
}
}
}
// TestCornerCases ensures several corner cases which can happen when opening
// a database and/or block files work as expected.
func TestCornerCases(t *testing.T) {
t.Parallel()
// Create a file at the datapase path to force the open below to fail.
dbPath := filepath.Join(os.TempDir(), "ffldb-errors")
_ = os.RemoveAll(dbPath)
fi, err := os.Create(dbPath)
if err != nil {
t.Errorf("os.Create: unexpected error: %v", err)
return
}
fi.Close()
// Ensure creating a new database fails when a file exists where a
// directory is needed.
testName := "openDB: fail due to file at target location"
wantErrCode := database.ErrDriverSpecific
idb, err := openDB(dbPath, blockDataNet, true)
if !checkDbError(t, testName, err, wantErrCode) {
if err == nil {
idb.Close()
}
_ = os.RemoveAll(dbPath)
return
}
// Remove the file and create the database to run tests against. It
// should be successful this time.
_ = os.RemoveAll(dbPath)
idb, err = openDB(dbPath, blockDataNet, true)
if err != nil {
t.Errorf("openDB: unexpected error: %v", err)
return
}
defer os.RemoveAll(dbPath)
defer idb.Close()
// Ensure attempting to write to a file that can't be created returns
// the expected error.
testName = "writeBlock: open file failure"
filePath := blockFilePath(dbPath, 0)
if err := os.Mkdir(filePath, 0755); err != nil {
t.Errorf("os.Mkdir: unexpected error: %v", err)
return
}
store := idb.(*db).store
_, err = store.writeBlock([]byte{0x00})
if !checkDbError(t, testName, err, database.ErrDriverSpecific) {
return
}
_ = os.RemoveAll(filePath)
// Close the underlying leveldb database out from under the database.
ldb := idb.(*db).cache.ldb
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
ldb.Close()
// Ensure initilization errors in the underlying database work as
// expected.
testName = "initDB: reinitialization"
wantErrCode = database.ErrDbNotOpen
err = initDB(ldb)
if !checkDbError(t, testName, err, wantErrCode) {
return
}
// Ensure the View handles errors in the underlying leveldb database
// properly.
testName = "View: underlying leveldb error"
wantErrCode = database.ErrDbNotOpen
err = idb.View(func(tx database.Tx) error {
return nil
})
if !checkDbError(t, testName, err, wantErrCode) {
return
}
// Ensure the Update handles errors in the underlying leveldb database
// properly.
testName = "Update: underlying leveldb error"
err = idb.Update(func(tx database.Tx) error {
return nil
})
if !checkDbError(t, testName, err, wantErrCode) {
return
}
}
// resetDatabase removes everything from the opened database associated with the
// test context including all metadata and the mock files.
func resetDatabase(tc *testContext) bool {
// Reset the metadata.
err := tc.db.Update(func(tx database.Tx) error {
// Remove all the keys using a cursor while also generating a
// list of buckets. It's not safe to remove keys during ForEach
// iteration nor is it safe to remove buckets during cursor
// iteration, so this dual approach is needed.
var bucketNames [][]byte
cursor := tx.Metadata().Cursor()
for ok := cursor.First(); ok; ok = cursor.Next() {
if cursor.Value() != nil {
if err := cursor.Delete(); err != nil {
return err
}
} else {
bucketNames = append(bucketNames, cursor.Key())
}
}
// Remove the buckets.
for _, k := range bucketNames {
if err := tx.Metadata().DeleteBucket(k); err != nil {
return err
}
}
_, err := tx.Metadata().CreateBucket(blockIdxBucketName)
return err
})
if err != nil {
tc.t.Errorf("Update: unexpected error: %v", err)
return false
}
// Reset the mock files.
store := tc.db.(*db).store
wc := store.writeCursor
wc.curFile.Lock()
if wc.curFile.file != nil {
wc.curFile.file.Close()
wc.curFile.file = nil
}
wc.curFile.Unlock()
wc.Lock()
wc.curFileNum = 0
wc.curOffset = 0
wc.Unlock()
tc.files = make(map[uint32]*lockableFile)
tc.maxFileSizes = make(map[uint32]int64)
return true
}
// testWriteFailures tests various failures paths when writing to the block
// files.
func testWriteFailures(tc *testContext) bool {
if !resetDatabase(tc) {
return false
}
// Ensure file sync errors during flush return the expected error.
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
store := tc.db.(*db).store
testName := "flush: file sync failure"
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
store.writeCursor.Lock()
oldFile := store.writeCursor.curFile
store.writeCursor.curFile = &lockableFile{
file: &mockFile{forceSyncErr: true, maxSize: -1},
}
store.writeCursor.Unlock()
err := tc.db.(*db).cache.flush()
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
if !checkDbError(tc.t, testName, err, database.ErrDriverSpecific) {
return false
}
store.writeCursor.Lock()
store.writeCursor.curFile = oldFile
store.writeCursor.Unlock()
// Force errors in the various error paths when writing data by using
// mock files with a limited max size.
block0Bytes, _ := tc.blocks[0].Bytes()
tests := []struct {
fileNum uint32
maxSize int64
}{
// Force an error when writing the network bytes.
{fileNum: 0, maxSize: 2},
// Force an error when writing the block size.
{fileNum: 0, maxSize: 6},
// Force an error when writing the block.
{fileNum: 0, maxSize: 17},
// Force an error when writing the checksum.
{fileNum: 0, maxSize: int64(len(block0Bytes)) + 10},
// Force an error after writing enough blocks for force multiple
// files.
{fileNum: 15, maxSize: 1},
}
for i, test := range tests {
if !resetDatabase(tc) {
return false
}
// Ensure storing the specified number of blocks using a mock
// file that fails the write fails when the transaction is
// committed, not when the block is stored.
tc.maxFileSizes = map[uint32]int64{test.fileNum: test.maxSize}
err := tc.db.Update(func(tx database.Tx) error {
for i, block := range tc.blocks {
err := tx.StoreBlock(block)
if err != nil {
tc.t.Errorf("StoreBlock (%d): unexpected "+
"error: %v", i, err)
return errSubTestFail
}
}
return nil
})
testName := fmt.Sprintf("Force update commit failure - test "+
"%d, fileNum %d, maxsize %d", i, test.fileNum,
test.maxSize)
if !checkDbError(tc.t, testName, err, database.ErrDriverSpecific) {
tc.t.Errorf("%v", err)
return false
}
// Ensure the commit rollback removed all extra files and data.
if len(tc.files) != 1 {
tc.t.Errorf("Update rollback: new not removed - want "+
"1 file, got %d", len(tc.files))
return false
}
if _, ok := tc.files[0]; !ok {
tc.t.Error("Update rollback: file 0 does not exist")
return false
}
file := tc.files[0].file.(*mockFile)
if len(file.data) != 0 {
tc.t.Errorf("Update rollback: file did not truncate - "+
"want len 0, got len %d", len(file.data))
return false
}
}
return true
}
// testBlockFileErrors ensures the database returns expected errors with various
// file-related issues such as closed and missing files.
func testBlockFileErrors(tc *testContext) bool {
if !resetDatabase(tc) {
return false
}
// Ensure errors in blockFile and openFile when requesting invalid file
// numbers.
store := tc.db.(*db).store
testName := "blockFile invalid file open"
_, err := store.blockFile(^uint32(0))
if !checkDbError(tc.t, testName, err, database.ErrDriverSpecific) {
return false
}
testName = "openFile invalid file open"
_, err = store.openFile(^uint32(0))
if !checkDbError(tc.t, testName, err, database.ErrDriverSpecific) {
return false
}
// Insert the first block into the mock file.
err = tc.db.Update(func(tx database.Tx) error {
err := tx.StoreBlock(tc.blocks[0])
if err != nil {
tc.t.Errorf("StoreBlock: unexpected error: %v", err)
return errSubTestFail
}
return nil
})
if err != nil {
if err != errSubTestFail {
tc.t.Errorf("Update: unexpected error: %v", err)
}
return false
}
// Ensure errors in readBlock and readBlockRegion when requesting a file
// number that doesn't exist.
block0Hash := tc.blocks[0].Hash()
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
testName = "readBlock invalid file number"
invalidLoc := blockLocation{
blockFileNum: ^uint32(0),
blockLen: 80,
}
_, err = store.readBlock(block0Hash, invalidLoc)
if !checkDbError(tc.t, testName, err, database.ErrDriverSpecific) {
return false
}
testName = "readBlockRegion invalid file number"
_, err = store.readBlockRegion(invalidLoc, 0, 80)
if !checkDbError(tc.t, testName, err, database.ErrDriverSpecific) {
return false
}
// Close the block file out from under the database.
store.writeCursor.curFile.Lock()
store.writeCursor.curFile.file.Close()
store.writeCursor.curFile.Unlock()
// Ensure failures in FetchBlock and FetchBlockRegion(s) since the
// underlying file they need to read from has been closed.
err = tc.db.View(func(tx database.Tx) error {
testName = "FetchBlock closed file"
wantErrCode := database.ErrDriverSpecific
_, err := tx.FetchBlock(block0Hash)
if !checkDbError(tc.t, testName, err, wantErrCode) {
return errSubTestFail
}
testName = "FetchBlockRegion closed file"
regions := []database.BlockRegion{
{
Hash: block0Hash,
Len: 80,
Offset: 0,
},
}
_, err = tx.FetchBlockRegion(&regions[0])
if !checkDbError(tc.t, testName, err, wantErrCode) {
return errSubTestFail
}
testName = "FetchBlockRegions closed file"
_, err = tx.FetchBlockRegions(regions)
if !checkDbError(tc.t, testName, err, wantErrCode) {
return errSubTestFail
}
return nil
})
if err != nil {
if err != errSubTestFail {
tc.t.Errorf("View: unexpected error: %v", err)
}
return false
}
return true
}
// testCorruption ensures the database returns expected errors under various
// corruption scenarios.
func testCorruption(tc *testContext) bool {
if !resetDatabase(tc) {
return false
}
// Insert the first block into the mock file.
err := tc.db.Update(func(tx database.Tx) error {
err := tx.StoreBlock(tc.blocks[0])
if err != nil {
tc.t.Errorf("StoreBlock: unexpected error: %v", err)
return errSubTestFail
}
return nil
})
if err != nil {
if err != errSubTestFail {
tc.t.Errorf("Update: unexpected error: %v", err)
}
return false
}
// Ensure corruption is detected by intentionally modifying the bytes
// stored to the mock file and reading the block.
block0Bytes, _ := tc.blocks[0].Bytes()
block0Hash := tc.blocks[0].Hash()
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
tests := []struct {
offset uint32
fixChecksum bool
wantErrCode database.ErrorCode
}{
// One of the network bytes. The checksum needs to be fixed so
// the invalid network is detected.
{2, true, database.ErrDriverSpecific},
// The same network byte, but this time don't fix the checksum
// to ensure the corruption is detected.
{2, false, database.ErrCorruption},
// One of the block length bytes.
{6, false, database.ErrCorruption},
// Random header byte.
{17, false, database.ErrCorruption},
// Random transaction byte.
{90, false, database.ErrCorruption},
// Random checksum byte.
{uint32(len(block0Bytes)) + 10, false, database.ErrCorruption},
}
err = tc.db.View(func(tx database.Tx) error {
data := tc.files[0].file.(*mockFile).data
for i, test := range tests {
// Corrupt the byte at the offset by a single bit.
data[test.offset] ^= 0x10
// Fix the checksum if requested to force other errors.
fileLen := len(data)
var oldChecksumBytes [4]byte
copy(oldChecksumBytes[:], data[fileLen-4:])
if test.fixChecksum {
toSum := data[:fileLen-4]
cksum := crc32.Checksum(toSum, castagnoli)
binary.BigEndian.PutUint32(data[fileLen-4:], cksum)
}
testName := fmt.Sprintf("FetchBlock (test #%d): "+
"corruption", i)
_, err := tx.FetchBlock(block0Hash)
if !checkDbError(tc.t, testName, err, test.wantErrCode) {
return errSubTestFail
}
// Reset the corrupted data back to the original.
data[test.offset] ^= 0x10
if test.fixChecksum {
copy(data[fileLen-4:], oldChecksumBytes[:])
}
}
return nil
})
if err != nil {
if err != errSubTestFail {
tc.t.Errorf("View: unexpected error: %v", err)
}
return false
}
return true
}
// TestFailureScenarios ensures several failure scenarios such as database
// corruption, block file write failures, and rollback failures are handled
// correctly.
func TestFailureScenarios(t *testing.T) {
// Create a new database to run tests against.
dbPath := filepath.Join(os.TempDir(), "ffldb-failurescenarios")
_ = os.RemoveAll(dbPath)
idb, err := database.Create(dbType, dbPath, blockDataNet)
if err != nil {
t.Errorf("Failed to create test database (%s) %v", dbType, err)
return
}
defer os.RemoveAll(dbPath)
defer idb.Close()
// Create a test context to pass around.
tc := &testContext{
t: t,
db: idb,
files: make(map[uint32]*lockableFile),
maxFileSizes: make(map[uint32]int64),
}
// Change the maximum file size to a small value to force multiple flat
// files with the test data set and replace the file-related functions
// to make use of mock files in memory. This allows injection of
// various file-related errors.
store := idb.(*db).store
store.maxBlockFileSize = 1024 // 1KiB
store.openWriteFileFunc = func(fileNum uint32) (filer, error) {
if file, ok := tc.files[fileNum]; ok {
// "Reopen" the file.
file.Lock()
mock := file.file.(*mockFile)
mock.Lock()
mock.closed = false
mock.Unlock()
file.Unlock()
return mock, nil
}
// Limit the max size of the mock file as specified in the test
// context.
maxSize := int64(-1)
if maxFileSize, ok := tc.maxFileSizes[fileNum]; ok {
2020-05-13 14:58:39 +02:00
maxSize = maxFileSize
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
}
2020-05-13 14:58:39 +02:00
file := &mockFile{maxSize: maxSize}
database: Major redesign of database package. This commit contains a complete redesign and rewrite of the database package that approaches things in a vastly different manner than the previous version. This is the first part of several stages that will be needed to ultimately make use of this new package. Some of the reason for this were discussed in #255, however a quick summary is as follows: - The previous database could only contain blocks on the main chain and reorgs required deleting the blocks from the database. This made it impossible to store orphans and could make external RPC calls for information about blocks during the middle of a reorg fail. - The previous database interface forced a high level of bitcoin-specific intelligence such as spend tracking into each backend driver. - The aforementioned point led to making it difficult to implement new backend drivers due to the need to repeat a lot of non-trivial logic which is better handled at a higher layer, such as the blockchain package. - The old database stored all blocks in leveldb. This made it extremely inefficient to do things such as lookup headers and individual transactions since the entire block had to be loaded from leveldb (which entails it doing data copies) to get access. In order to address all of these concerns, and others not mentioned, the database interface has been redesigned as follows: - Two main categories of functionality are provided: block storage and metadata storage - All block storage and metadata storage are done via read-only and read-write MVCC transactions with both manual and managed modes - Support for multiple concurrent readers and a single writer - Readers use a snapshot and therefore are not blocked by the writer - Some key properties of the block storage and retrieval API: - It is generic and does NOT contain additional bitcoin logic such spend tracking and block linking - Provides access to the raw serialized bytes so deserialization is not forced for callers that don't need it - Support for fetching headers via independent functions which allows implementations to provide significant optimizations - Ability to efficiently retrieve arbitrary regions of blocks (transactions, scripts, etc) - A rich metadata storage API is provided: - Key/value with arbitrary data - Support for buckets and nested buckets - Bucket iteration through a couple of different mechanisms - Cursors for efficient and direct key seeking - Supports registration of backend database implementations - Comprehensive test coverage - Provides strong documentation with example usage This commit also contains an implementation of the previously discussed interface named ffldb (flat file plus leveldb metadata backend). Here is a quick overview: - Highly optimized for read performance with consistent write performance regardless of database size - All blocks are stored in flat files on the file system - Bulk block region fetching is optimized to perform linear reads which improves performance on spindle disks - Anti-corruption mechanisms: - Flat files contain full block checksums to quickly an easily detect database corruption without needing to do expensive merkle root calculations - Metadata checksums - Open reconciliation - Extensive test coverage: - Comprehensive blackbox interface testing - Whitebox testing which uses intimate knowledge to exercise uncommon failure paths such as deleting files out from under the database - Corruption tests (replacing random data in the files) In addition, this commit also contains a new tool under the new database directory named dbtool which provides a few basic commands for testing the database. It is designed around commands, so it could be useful to expand on in the future. Finally, this commit addresses the following issues: - Adds support for and therefore closes #255 - Fixes #199 - Fixes #201 - Implements and closes #256 - Obsoletes and closes #257 - Closes #247 once the required chain and btcd modifications are in place to make use of this new code
2016-02-03 18:42:04 +01:00
tc.files[fileNum] = &lockableFile{file: file}
return file, nil
}
store.openFileFunc = func(fileNum uint32) (*lockableFile, error) {
// Force error when trying to open max file num.
if fileNum == ^uint32(0) {
return nil, makeDbErr(database.ErrDriverSpecific,
"test", nil)
}
if file, ok := tc.files[fileNum]; ok {
// "Reopen" the file.
file.Lock()
mock := file.file.(*mockFile)
mock.Lock()
mock.closed = false
mock.Unlock()
file.Unlock()
return file, nil
}
file := &lockableFile{file: &mockFile{}}
tc.files[fileNum] = file
return file, nil
}
store.deleteFileFunc = func(fileNum uint32) error {
if file, ok := tc.files[fileNum]; ok {
file.Lock()
file.file.Close()
file.Unlock()
delete(tc.files, fileNum)
return nil
}
str := fmt.Sprintf("file %d does not exist", fileNum)
return makeDbErr(database.ErrDriverSpecific, str, nil)
}
// Load the test blocks and save in the test context for use throughout
// the tests.
blocks, err := loadBlocks(t, blockDataFile, blockDataNet)
if err != nil {
t.Errorf("loadBlocks: Unexpected error: %v", err)
return
}
tc.blocks = blocks
// Test various failures paths when writing to the block files.
if !testWriteFailures(tc) {
return
}
// Test various file-related issues such as closed and missing files.
if !testBlockFileErrors(tc) {
return
}
// Test various corruption scenarios.
testCorruption(tc)
}