diff options
author | Péter Szilágyi <peterke@gmail.com> | 2019-05-17 00:10:58 +0800 |
---|---|---|
committer | GitHub <noreply@github.com> | 2019-05-17 00:10:58 +0800 |
commit | f5d89cdb72c1e82e9deb54754bef8dd20bf12591 (patch) | |
tree | b6e2ee16b90cd494eaefec59fb4aaaac71cc6a96 /core/rawdb | |
parent | 60386b3545659d99c7488e456c780606db101936 (diff) | |
parent | 9eba3a9fff2f47f5e094c36a7c905380b0ac8b1f (diff) | |
download | go-tangerine-f5d89cdb72c1e82e9deb54754bef8dd20bf12591.tar go-tangerine-f5d89cdb72c1e82e9deb54754bef8dd20bf12591.tar.gz go-tangerine-f5d89cdb72c1e82e9deb54754bef8dd20bf12591.tar.bz2 go-tangerine-f5d89cdb72c1e82e9deb54754bef8dd20bf12591.tar.lz go-tangerine-f5d89cdb72c1e82e9deb54754bef8dd20bf12591.tar.xz go-tangerine-f5d89cdb72c1e82e9deb54754bef8dd20bf12591.tar.zst go-tangerine-f5d89cdb72c1e82e9deb54754bef8dd20bf12591.zip |
Merge pull request #19244 from karalabe/freezer-2
cmd, core, eth, les, node: chain freezer on top of db rework
Diffstat (limited to 'core/rawdb')
-rw-r--r-- | core/rawdb/accessors_chain.go | 198 | ||||
-rw-r--r-- | core/rawdb/accessors_indexes.go | 8 | ||||
-rw-r--r-- | core/rawdb/accessors_metadata.go | 12 | ||||
-rw-r--r-- | core/rawdb/database.go | 311 | ||||
-rw-r--r-- | core/rawdb/freezer.go | 382 | ||||
-rw-r--r-- | core/rawdb/freezer_table.go | 561 | ||||
-rw-r--r-- | core/rawdb/freezer_table_test.go | 609 | ||||
-rw-r--r-- | core/rawdb/schema.go | 27 | ||||
-rw-r--r-- | core/rawdb/table.go | 44 |
9 files changed, 2098 insertions, 54 deletions
diff --git a/core/rawdb/accessors_chain.go b/core/rawdb/accessors_chain.go index cc0591a4c..fab7ca56c 100644 --- a/core/rawdb/accessors_chain.go +++ b/core/rawdb/accessors_chain.go @@ -31,7 +31,17 @@ import ( // ReadCanonicalHash retrieves the hash assigned to a canonical block number. func ReadCanonicalHash(db ethdb.Reader, number uint64) common.Hash { - data, _ := db.Get(headerHashKey(number)) + data, _ := db.Ancient(freezerHashTable, number) + if len(data) == 0 { + data, _ = db.Get(headerHashKey(number)) + // In the background freezer is moving data from leveldb to flatten files. + // So during the first check for ancient db, the data is not yet in there, + // but when we reach into leveldb, the data was already moved. That would + // result in a not found error. + if len(data) == 0 { + data, _ = db.Ancient(freezerHashTable, number) + } + } if len(data) == 0 { return common.Hash{} } @@ -39,21 +49,38 @@ func ReadCanonicalHash(db ethdb.Reader, number uint64) common.Hash { } // WriteCanonicalHash stores the hash assigned to a canonical block number. -func WriteCanonicalHash(db ethdb.Writer, hash common.Hash, number uint64) { +func WriteCanonicalHash(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { if err := db.Put(headerHashKey(number), hash.Bytes()); err != nil { log.Crit("Failed to store number to hash mapping", "err", err) } } // DeleteCanonicalHash removes the number to hash canonical mapping. -func DeleteCanonicalHash(db ethdb.Writer, number uint64) { +func DeleteCanonicalHash(db ethdb.KeyValueWriter, number uint64) { if err := db.Delete(headerHashKey(number)); err != nil { log.Crit("Failed to delete number to hash mapping", "err", err) } } +// ReadAllHashes retrieves all the hashes assigned to blocks at a certain heights, +// both canonical and reorged forks included. +func ReadAllHashes(db ethdb.Iteratee, number uint64) []common.Hash { + prefix := headerKeyPrefix(number) + + hashes := make([]common.Hash, 0, 1) + it := db.NewIteratorWithPrefix(prefix) + defer it.Release() + + for it.Next() { + if key := it.Key(); len(key) == len(prefix)+32 { + hashes = append(hashes, common.BytesToHash(key[len(key)-32:])) + } + } + return hashes +} + // ReadHeaderNumber returns the header number assigned to a hash. -func ReadHeaderNumber(db ethdb.Reader, hash common.Hash) *uint64 { +func ReadHeaderNumber(db ethdb.KeyValueReader, hash common.Hash) *uint64 { data, _ := db.Get(headerNumberKey(hash)) if len(data) != 8 { return nil @@ -62,8 +89,24 @@ func ReadHeaderNumber(db ethdb.Reader, hash common.Hash) *uint64 { return &number } +// WriteHeaderNumber stores the hash->number mapping. +func WriteHeaderNumber(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { + key := headerNumberKey(hash) + enc := encodeBlockNumber(number) + if err := db.Put(key, enc); err != nil { + log.Crit("Failed to store hash to number mapping", "err", err) + } +} + +// DeleteHeaderNumber removes hash->number mapping. +func DeleteHeaderNumber(db ethdb.KeyValueWriter, hash common.Hash) { + if err := db.Delete(headerNumberKey(hash)); err != nil { + log.Crit("Failed to delete hash to number mapping", "err", err) + } +} + // ReadHeadHeaderHash retrieves the hash of the current canonical head header. -func ReadHeadHeaderHash(db ethdb.Reader) common.Hash { +func ReadHeadHeaderHash(db ethdb.KeyValueReader) common.Hash { data, _ := db.Get(headHeaderKey) if len(data) == 0 { return common.Hash{} @@ -72,14 +115,14 @@ func ReadHeadHeaderHash(db ethdb.Reader) common.Hash { } // WriteHeadHeaderHash stores the hash of the current canonical head header. -func WriteHeadHeaderHash(db ethdb.Writer, hash common.Hash) { +func WriteHeadHeaderHash(db ethdb.KeyValueWriter, hash common.Hash) { if err := db.Put(headHeaderKey, hash.Bytes()); err != nil { log.Crit("Failed to store last header's hash", "err", err) } } // ReadHeadBlockHash retrieves the hash of the current canonical head block. -func ReadHeadBlockHash(db ethdb.Reader) common.Hash { +func ReadHeadBlockHash(db ethdb.KeyValueReader) common.Hash { data, _ := db.Get(headBlockKey) if len(data) == 0 { return common.Hash{} @@ -88,14 +131,14 @@ func ReadHeadBlockHash(db ethdb.Reader) common.Hash { } // WriteHeadBlockHash stores the head block's hash. -func WriteHeadBlockHash(db ethdb.Writer, hash common.Hash) { +func WriteHeadBlockHash(db ethdb.KeyValueWriter, hash common.Hash) { if err := db.Put(headBlockKey, hash.Bytes()); err != nil { log.Crit("Failed to store last block's hash", "err", err) } } // ReadHeadFastBlockHash retrieves the hash of the current fast-sync head block. -func ReadHeadFastBlockHash(db ethdb.Reader) common.Hash { +func ReadHeadFastBlockHash(db ethdb.KeyValueReader) common.Hash { data, _ := db.Get(headFastBlockKey) if len(data) == 0 { return common.Hash{} @@ -104,7 +147,7 @@ func ReadHeadFastBlockHash(db ethdb.Reader) common.Hash { } // WriteHeadFastBlockHash stores the hash of the current fast-sync head block. -func WriteHeadFastBlockHash(db ethdb.Writer, hash common.Hash) { +func WriteHeadFastBlockHash(db ethdb.KeyValueWriter, hash common.Hash) { if err := db.Put(headFastBlockKey, hash.Bytes()); err != nil { log.Crit("Failed to store last fast block's hash", "err", err) } @@ -112,7 +155,7 @@ func WriteHeadFastBlockHash(db ethdb.Writer, hash common.Hash) { // ReadFastTrieProgress retrieves the number of tries nodes fast synced to allow // reporting correct numbers across restarts. -func ReadFastTrieProgress(db ethdb.Reader) uint64 { +func ReadFastTrieProgress(db ethdb.KeyValueReader) uint64 { data, _ := db.Get(fastTrieProgressKey) if len(data) == 0 { return 0 @@ -122,7 +165,7 @@ func ReadFastTrieProgress(db ethdb.Reader) uint64 { // WriteFastTrieProgress stores the fast sync trie process counter to support // retrieving it across restarts. -func WriteFastTrieProgress(db ethdb.Writer, count uint64) { +func WriteFastTrieProgress(db ethdb.KeyValueWriter, count uint64) { if err := db.Put(fastTrieProgressKey, new(big.Int).SetUint64(count).Bytes()); err != nil { log.Crit("Failed to store fast sync trie progress", "err", err) } @@ -130,12 +173,25 @@ func WriteFastTrieProgress(db ethdb.Writer, count uint64) { // ReadHeaderRLP retrieves a block header in its raw RLP database encoding. func ReadHeaderRLP(db ethdb.Reader, hash common.Hash, number uint64) rlp.RawValue { - data, _ := db.Get(headerKey(number, hash)) + data, _ := db.Ancient(freezerHeaderTable, number) + if len(data) == 0 { + data, _ = db.Get(headerKey(number, hash)) + // In the background freezer is moving data from leveldb to flatten files. + // So during the first check for ancient db, the data is not yet in there, + // but when we reach into leveldb, the data was already moved. That would + // result in a not found error. + if len(data) == 0 { + data, _ = db.Ancient(freezerHeaderTable, number) + } + } return data } // HasHeader verifies the existence of a block header corresponding to the hash. func HasHeader(db ethdb.Reader, hash common.Hash, number uint64) bool { + if has, err := db.Ancient(freezerHashTable, number); err == nil && common.BytesToHash(has) == hash { + return true + } if has, err := db.Has(headerKey(number, hash)); !has || err != nil { return false } @@ -158,30 +214,27 @@ func ReadHeader(db ethdb.Reader, hash common.Hash, number uint64) *types.Header // WriteHeader stores a block header into the database and also stores the hash- // to-number mapping. -func WriteHeader(db ethdb.Writer, header *types.Header) { - // Write the hash -> number mapping +func WriteHeader(db ethdb.KeyValueWriter, header *types.Header) { var ( - hash = header.Hash() - number = header.Number.Uint64() - encoded = encodeBlockNumber(number) + hash = header.Hash() + number = header.Number.Uint64() ) - key := headerNumberKey(hash) - if err := db.Put(key, encoded); err != nil { - log.Crit("Failed to store hash to number mapping", "err", err) - } + // Write the hash -> number mapping + WriteHeaderNumber(db, hash, number) + // Write the encoded header data, err := rlp.EncodeToBytes(header) if err != nil { log.Crit("Failed to RLP encode header", "err", err) } - key = headerKey(number, hash) + key := headerKey(number, hash) if err := db.Put(key, data); err != nil { log.Crit("Failed to store header", "err", err) } } // DeleteHeader removes all block header data associated with a hash. -func DeleteHeader(db ethdb.Writer, hash common.Hash, number uint64) { +func DeleteHeader(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { deleteHeaderWithoutNumber(db, hash, number) if err := db.Delete(headerNumberKey(hash)); err != nil { log.Crit("Failed to delete hash to number mapping", "err", err) @@ -190,7 +243,7 @@ func DeleteHeader(db ethdb.Writer, hash common.Hash, number uint64) { // deleteHeaderWithoutNumber removes only the block header but does not remove // the hash to number mapping. -func deleteHeaderWithoutNumber(db ethdb.Writer, hash common.Hash, number uint64) { +func deleteHeaderWithoutNumber(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { if err := db.Delete(headerKey(number, hash)); err != nil { log.Crit("Failed to delete header", "err", err) } @@ -198,12 +251,22 @@ func deleteHeaderWithoutNumber(db ethdb.Writer, hash common.Hash, number uint64) // ReadBodyRLP retrieves the block body (transactions and uncles) in RLP encoding. func ReadBodyRLP(db ethdb.Reader, hash common.Hash, number uint64) rlp.RawValue { - data, _ := db.Get(blockBodyKey(number, hash)) + data, _ := db.Ancient(freezerBodiesTable, number) + if len(data) == 0 { + data, _ = db.Get(blockBodyKey(number, hash)) + // In the background freezer is moving data from leveldb to flatten files. + // So during the first check for ancient db, the data is not yet in there, + // but when we reach into leveldb, the data was already moved. That would + // result in a not found error. + if len(data) == 0 { + data, _ = db.Ancient(freezerBodiesTable, number) + } + } return data } // WriteBodyRLP stores an RLP encoded block body into the database. -func WriteBodyRLP(db ethdb.Writer, hash common.Hash, number uint64, rlp rlp.RawValue) { +func WriteBodyRLP(db ethdb.KeyValueWriter, hash common.Hash, number uint64, rlp rlp.RawValue) { if err := db.Put(blockBodyKey(number, hash), rlp); err != nil { log.Crit("Failed to store block body", "err", err) } @@ -211,6 +274,9 @@ func WriteBodyRLP(db ethdb.Writer, hash common.Hash, number uint64, rlp rlp.RawV // HasBody verifies the existence of a block body corresponding to the hash. func HasBody(db ethdb.Reader, hash common.Hash, number uint64) bool { + if has, err := db.Ancient(freezerHashTable, number); err == nil && common.BytesToHash(has) == hash { + return true + } if has, err := db.Has(blockBodyKey(number, hash)); !has || err != nil { return false } @@ -232,7 +298,7 @@ func ReadBody(db ethdb.Reader, hash common.Hash, number uint64) *types.Body { } // WriteBody stores a block body into the database. -func WriteBody(db ethdb.Writer, hash common.Hash, number uint64, body *types.Body) { +func WriteBody(db ethdb.KeyValueWriter, hash common.Hash, number uint64, body *types.Body) { data, err := rlp.EncodeToBytes(body) if err != nil { log.Crit("Failed to RLP encode body", "err", err) @@ -241,7 +307,7 @@ func WriteBody(db ethdb.Writer, hash common.Hash, number uint64, body *types.Bod } // DeleteBody removes all block body data associated with a hash. -func DeleteBody(db ethdb.Writer, hash common.Hash, number uint64) { +func DeleteBody(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { if err := db.Delete(blockBodyKey(number, hash)); err != nil { log.Crit("Failed to delete block body", "err", err) } @@ -249,7 +315,17 @@ func DeleteBody(db ethdb.Writer, hash common.Hash, number uint64) { // ReadTdRLP retrieves a block's total difficulty corresponding to the hash in RLP encoding. func ReadTdRLP(db ethdb.Reader, hash common.Hash, number uint64) rlp.RawValue { - data, _ := db.Get(headerTDKey(number, hash)) + data, _ := db.Ancient(freezerDifficultyTable, number) + if len(data) == 0 { + data, _ = db.Get(headerTDKey(number, hash)) + // In the background freezer is moving data from leveldb to flatten files. + // So during the first check for ancient db, the data is not yet in there, + // but when we reach into leveldb, the data was already moved. That would + // result in a not found error. + if len(data) == 0 { + data, _ = db.Ancient(freezerDifficultyTable, number) + } + } return data } @@ -268,7 +344,7 @@ func ReadTd(db ethdb.Reader, hash common.Hash, number uint64) *big.Int { } // WriteTd stores the total difficulty of a block into the database. -func WriteTd(db ethdb.Writer, hash common.Hash, number uint64, td *big.Int) { +func WriteTd(db ethdb.KeyValueWriter, hash common.Hash, number uint64, td *big.Int) { data, err := rlp.EncodeToBytes(td) if err != nil { log.Crit("Failed to RLP encode block total difficulty", "err", err) @@ -279,7 +355,7 @@ func WriteTd(db ethdb.Writer, hash common.Hash, number uint64, td *big.Int) { } // DeleteTd removes all block total difficulty data associated with a hash. -func DeleteTd(db ethdb.Writer, hash common.Hash, number uint64) { +func DeleteTd(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { if err := db.Delete(headerTDKey(number, hash)); err != nil { log.Crit("Failed to delete block total difficulty", "err", err) } @@ -288,6 +364,9 @@ func DeleteTd(db ethdb.Writer, hash common.Hash, number uint64) { // HasReceipts verifies the existence of all the transaction receipts belonging // to a block. func HasReceipts(db ethdb.Reader, hash common.Hash, number uint64) bool { + if has, err := db.Ancient(freezerHashTable, number); err == nil && common.BytesToHash(has) == hash { + return true + } if has, err := db.Has(blockReceiptsKey(number, hash)); !has || err != nil { return false } @@ -296,7 +375,17 @@ func HasReceipts(db ethdb.Reader, hash common.Hash, number uint64) bool { // ReadReceiptsRLP retrieves all the transaction receipts belonging to a block in RLP encoding. func ReadReceiptsRLP(db ethdb.Reader, hash common.Hash, number uint64) rlp.RawValue { - data, _ := db.Get(blockReceiptsKey(number, hash)) + data, _ := db.Ancient(freezerReceiptTable, number) + if len(data) == 0 { + data, _ = db.Get(blockReceiptsKey(number, hash)) + // In the background freezer is moving data from leveldb to flatten files. + // So during the first check for ancient db, the data is not yet in there, + // but when we reach into leveldb, the data was already moved. That would + // result in a not found error. + if len(data) == 0 { + data, _ = db.Ancient(freezerReceiptTable, number) + } + } return data } @@ -348,7 +437,7 @@ func ReadReceipts(db ethdb.Reader, hash common.Hash, number uint64, config *para } // WriteReceipts stores all the transaction receipts belonging to a block. -func WriteReceipts(db ethdb.Writer, hash common.Hash, number uint64, receipts types.Receipts) { +func WriteReceipts(db ethdb.KeyValueWriter, hash common.Hash, number uint64, receipts types.Receipts) { // Convert the receipts into their storage form and serialize them storageReceipts := make([]*types.ReceiptForStorage, len(receipts)) for i, receipt := range receipts { @@ -365,7 +454,7 @@ func WriteReceipts(db ethdb.Writer, hash common.Hash, number uint64, receipts ty } // DeleteReceipts removes all receipt data associated with a block hash. -func DeleteReceipts(db ethdb.Writer, hash common.Hash, number uint64) { +func DeleteReceipts(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { if err := db.Delete(blockReceiptsKey(number, hash)); err != nil { log.Crit("Failed to delete block receipts", "err", err) } @@ -390,22 +479,53 @@ func ReadBlock(db ethdb.Reader, hash common.Hash, number uint64) *types.Block { } // WriteBlock serializes a block into the database, header and body separately. -func WriteBlock(db ethdb.Writer, block *types.Block) { +func WriteBlock(db ethdb.KeyValueWriter, block *types.Block) { WriteBody(db, block.Hash(), block.NumberU64(), block.Body()) WriteHeader(db, block.Header()) } +// WriteAncientBlock writes entire block data into ancient store and returns the total written size. +func WriteAncientBlock(db ethdb.AncientWriter, block *types.Block, receipts types.Receipts, td *big.Int) int { + // Encode all block components to RLP format. + headerBlob, err := rlp.EncodeToBytes(block.Header()) + if err != nil { + log.Crit("Failed to RLP encode block header", "err", err) + } + bodyBlob, err := rlp.EncodeToBytes(block.Body()) + if err != nil { + log.Crit("Failed to RLP encode body", "err", err) + } + storageReceipts := make([]*types.ReceiptForStorage, len(receipts)) + for i, receipt := range receipts { + storageReceipts[i] = (*types.ReceiptForStorage)(receipt) + } + receiptBlob, err := rlp.EncodeToBytes(storageReceipts) + if err != nil { + log.Crit("Failed to RLP encode block receipts", "err", err) + } + tdBlob, err := rlp.EncodeToBytes(td) + if err != nil { + log.Crit("Failed to RLP encode block total difficulty", "err", err) + } + // Write all blob to flatten files. + err = db.AppendAncient(block.NumberU64(), block.Hash().Bytes(), headerBlob, bodyBlob, receiptBlob, tdBlob) + if err != nil { + log.Crit("Failed to write block data to ancient store", "err", err) + } + return len(headerBlob) + len(bodyBlob) + len(receiptBlob) + len(tdBlob) + common.HashLength +} + // DeleteBlock removes all block data associated with a hash. -func DeleteBlock(db ethdb.Writer, hash common.Hash, number uint64) { +func DeleteBlock(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { DeleteReceipts(db, hash, number) DeleteHeader(db, hash, number) DeleteBody(db, hash, number) DeleteTd(db, hash, number) } -// deleteBlockWithoutNumber removes all block data associated with a hash, except +// DeleteBlockWithoutNumber removes all block data associated with a hash, except // the hash to number mapping. -func deleteBlockWithoutNumber(db ethdb.Writer, hash common.Hash, number uint64) { +func DeleteBlockWithoutNumber(db ethdb.KeyValueWriter, hash common.Hash, number uint64) { DeleteReceipts(db, hash, number) deleteHeaderWithoutNumber(db, hash, number) DeleteBody(db, hash, number) diff --git a/core/rawdb/accessors_indexes.go b/core/rawdb/accessors_indexes.go index 423145a76..ed1f1bca6 100644 --- a/core/rawdb/accessors_indexes.go +++ b/core/rawdb/accessors_indexes.go @@ -54,7 +54,7 @@ func ReadTxLookupEntry(db ethdb.Reader, hash common.Hash) *uint64 { // WriteTxLookupEntries stores a positional metadata for every transaction from // a block, enabling hash based transaction and receipt lookups. -func WriteTxLookupEntries(db ethdb.Writer, block *types.Block) { +func WriteTxLookupEntries(db ethdb.KeyValueWriter, block *types.Block) { for _, tx := range block.Transactions() { if err := db.Put(txLookupKey(tx.Hash()), block.Number().Bytes()); err != nil { log.Crit("Failed to store transaction lookup entry", "err", err) @@ -63,7 +63,7 @@ func WriteTxLookupEntries(db ethdb.Writer, block *types.Block) { } // DeleteTxLookupEntry removes all transaction data associated with a hash. -func DeleteTxLookupEntry(db ethdb.Writer, hash common.Hash) { +func DeleteTxLookupEntry(db ethdb.KeyValueWriter, hash common.Hash) { db.Delete(txLookupKey(hash)) } @@ -117,13 +117,13 @@ func ReadReceipt(db ethdb.Reader, hash common.Hash, config *params.ChainConfig) // ReadBloomBits retrieves the compressed bloom bit vector belonging to the given // section and bit index from the. -func ReadBloomBits(db ethdb.Reader, bit uint, section uint64, head common.Hash) ([]byte, error) { +func ReadBloomBits(db ethdb.KeyValueReader, bit uint, section uint64, head common.Hash) ([]byte, error) { return db.Get(bloomBitsKey(bit, section, head)) } // WriteBloomBits stores the compressed bloom bits vector belonging to the given // section and bit index. -func WriteBloomBits(db ethdb.Writer, bit uint, section uint64, head common.Hash, bits []byte) { +func WriteBloomBits(db ethdb.KeyValueWriter, bit uint, section uint64, head common.Hash, bits []byte) { if err := db.Put(bloomBitsKey(bit, section, head), bits); err != nil { log.Crit("Failed to store bloom bits", "err", err) } diff --git a/core/rawdb/accessors_metadata.go b/core/rawdb/accessors_metadata.go index 1361b0d73..f8d09fbdd 100644 --- a/core/rawdb/accessors_metadata.go +++ b/core/rawdb/accessors_metadata.go @@ -27,7 +27,7 @@ import ( ) // ReadDatabaseVersion retrieves the version number of the database. -func ReadDatabaseVersion(db ethdb.Reader) *uint64 { +func ReadDatabaseVersion(db ethdb.KeyValueReader) *uint64 { var version uint64 enc, _ := db.Get(databaseVerisionKey) @@ -42,7 +42,7 @@ func ReadDatabaseVersion(db ethdb.Reader) *uint64 { } // WriteDatabaseVersion stores the version number of the database -func WriteDatabaseVersion(db ethdb.Writer, version uint64) { +func WriteDatabaseVersion(db ethdb.KeyValueWriter, version uint64) { enc, err := rlp.EncodeToBytes(version) if err != nil { log.Crit("Failed to encode database version", "err", err) @@ -53,7 +53,7 @@ func WriteDatabaseVersion(db ethdb.Writer, version uint64) { } // ReadChainConfig retrieves the consensus settings based on the given genesis hash. -func ReadChainConfig(db ethdb.Reader, hash common.Hash) *params.ChainConfig { +func ReadChainConfig(db ethdb.KeyValueReader, hash common.Hash) *params.ChainConfig { data, _ := db.Get(configKey(hash)) if len(data) == 0 { return nil @@ -67,7 +67,7 @@ func ReadChainConfig(db ethdb.Reader, hash common.Hash) *params.ChainConfig { } // WriteChainConfig writes the chain config settings to the database. -func WriteChainConfig(db ethdb.Writer, hash common.Hash, cfg *params.ChainConfig) { +func WriteChainConfig(db ethdb.KeyValueWriter, hash common.Hash, cfg *params.ChainConfig) { if cfg == nil { return } @@ -81,13 +81,13 @@ func WriteChainConfig(db ethdb.Writer, hash common.Hash, cfg *params.ChainConfig } // ReadPreimage retrieves a single preimage of the provided hash. -func ReadPreimage(db ethdb.Reader, hash common.Hash) []byte { +func ReadPreimage(db ethdb.KeyValueReader, hash common.Hash) []byte { data, _ := db.Get(preimageKey(hash)) return data } // WritePreimages writes the provided set of preimages to the database. -func WritePreimages(db ethdb.Writer, preimages map[common.Hash][]byte) { +func WritePreimages(db ethdb.KeyValueWriter, preimages map[common.Hash][]byte) { for hash, preimage := range preimages { if err := db.Put(preimageKey(hash), preimage); err != nil { log.Crit("Failed to store trie preimage", "err", err) diff --git a/core/rawdb/database.go b/core/rawdb/database.go index b4c5dea70..353b7dce6 100644 --- a/core/rawdb/database.go +++ b/core/rawdb/database.go @@ -17,15 +17,168 @@ package rawdb import ( + "bytes" + "errors" + "fmt" + "os" + "time" + + "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/ethdb" "github.com/ethereum/go-ethereum/ethdb/leveldb" "github.com/ethereum/go-ethereum/ethdb/memorydb" + "github.com/ethereum/go-ethereum/log" + "github.com/olekukonko/tablewriter" ) +// freezerdb is a database wrapper that enabled freezer data retrievals. +type freezerdb struct { + ethdb.KeyValueStore + ethdb.AncientStore +} + +// Close implements io.Closer, closing both the fast key-value store as well as +// the slow ancient tables. +func (frdb *freezerdb) Close() error { + var errs []error + if err := frdb.KeyValueStore.Close(); err != nil { + errs = append(errs, err) + } + if err := frdb.AncientStore.Close(); err != nil { + errs = append(errs, err) + } + if len(errs) != 0 { + return fmt.Errorf("%v", errs) + } + return nil +} + +// nofreezedb is a database wrapper that disables freezer data retrievals. +type nofreezedb struct { + ethdb.KeyValueStore +} + +// HasAncient returns an error as we don't have a backing chain freezer. +func (db *nofreezedb) HasAncient(kind string, number uint64) (bool, error) { + return false, errNotSupported +} + +// Ancient returns an error as we don't have a backing chain freezer. +func (db *nofreezedb) Ancient(kind string, number uint64) ([]byte, error) { + return nil, errNotSupported +} + +// Ancients returns an error as we don't have a backing chain freezer. +func (db *nofreezedb) Ancients() (uint64, error) { + return 0, errNotSupported +} + +// AncientSize returns an error as we don't have a backing chain freezer. +func (db *nofreezedb) AncientSize(kind string) (uint64, error) { + return 0, errNotSupported +} + +// AppendAncient returns an error as we don't have a backing chain freezer. +func (db *nofreezedb) AppendAncient(number uint64, hash, header, body, receipts, td []byte) error { + return errNotSupported +} + +// TruncateAncients returns an error as we don't have a backing chain freezer. +func (db *nofreezedb) TruncateAncients(items uint64) error { + return errNotSupported +} + +// Sync returns an error as we don't have a backing chain freezer. +func (db *nofreezedb) Sync() error { + return errNotSupported +} + // NewDatabase creates a high level database on top of a given key-value data // store without a freezer moving immutable chain segments into cold storage. func NewDatabase(db ethdb.KeyValueStore) ethdb.Database { - return db + return &nofreezedb{ + KeyValueStore: db, + } +} + +// NewDatabaseWithFreezer creates a high level database on top of a given key- +// value data store with a freezer moving immutable chain segments into cold +// storage. +func NewDatabaseWithFreezer(db ethdb.KeyValueStore, freezer string, namespace string) (ethdb.Database, error) { + // Create the idle freezer instance + frdb, err := newFreezer(freezer, namespace) + if err != nil { + return nil, err + } + // Since the freezer can be stored separately from the user's key-value database, + // there's a fairly high probability that the user requests invalid combinations + // of the freezer and database. Ensure that we don't shoot ourselves in the foot + // by serving up conflicting data, leading to both datastores getting corrupted. + // + // - If both the freezer and key-value store is empty (no genesis), we just + // initialized a new empty freezer, so everything's fine. + // - If the key-value store is empty, but the freezer is not, we need to make + // sure the user's genesis matches the freezer. That will be checked in the + // blockchain, since we don't have the genesis block here (nor should we at + // this point care, the key-value/freezer combo is valid). + // - If neither the key-value store nor the freezer is empty, cross validate + // the genesis hashes to make sure they are compatible. If they are, also + // ensure that there's no gap between the freezer and sunsequently leveldb. + // - If the key-value store is not empty, but the freezer is we might just be + // upgrading to the freezer release, or we might have had a small chain and + // not frozen anything yet. Ensure that no blocks are missing yet from the + // key-value store, since that would mean we already had an old freezer. + + // If the genesis hash is empty, we have a new key-value store, so nothing to + // validate in this method. If, however, the genesis hash is not nil, compare + // it to the freezer content. + if kvgenesis, _ := db.Get(headerHashKey(0)); len(kvgenesis) > 0 { + if frozen, _ := frdb.Ancients(); frozen > 0 { + // If the freezer already contains something, ensure that the genesis blocks + // match, otherwise we might mix up freezers across chains and destroy both + // the freezer and the key-value store. + if frgenesis, _ := frdb.Ancient(freezerHashTable, 0); !bytes.Equal(kvgenesis, frgenesis) { + return nil, fmt.Errorf("genesis mismatch: %#x (leveldb) != %#x (ancients)", kvgenesis, frgenesis) + } + // Key-value store and freezer belong to the same network. Ensure that they + // are contiguous, otherwise we might end up with a non-functional freezer. + if kvhash, _ := db.Get(headerHashKey(frozen)); len(kvhash) == 0 { + // Subsequent header after the freezer limit is missing from the database. + // Reject startup is the database has a more recent head. + if *ReadHeaderNumber(db, ReadHeadHeaderHash(db)) > frozen-1 { + return nil, fmt.Errorf("gap (#%d) in the chain between ancients and leveldb", frozen) + } + // Database contains only older data than the freezer, this happens if the + // state was wiped and reinited from an existing freezer. + } else { + // Key-value store continues where the freezer left off, all is fine. We might + // have duplicate blocks (crash after freezer write but before kay-value store + // deletion, but that's fine). + } + } else { + // If the freezer is empty, ensure nothing was moved yet from the key-value + // store, otherwise we'll end up missing data. We check block #1 to decide + // if we froze anything previously or not, but do take care of databases with + // only the genesis block. + if ReadHeadHeaderHash(db) != common.BytesToHash(kvgenesis) { + // Key-value store contains more data than the genesis block, make sure we + // didn't freeze anything yet. + if kvblob, _ := db.Get(headerHashKey(1)); len(kvblob) == 0 { + return nil, errors.New("ancient chain segments already extracted, please set --datadir.ancient to the correct path") + } + // Block #1 is still in the database, we're allowed to init a new feezer + } else { + // The head header is still the genesis, we're allowed to init a new feezer + } + } + } + // Freezer is consistent with the key-value database, permit combining the two + go frdb.freeze(db) + + return &freezerdb{ + KeyValueStore: db, + AncientStore: frdb, + }, nil } // NewMemoryDatabase creates an ephemeral in-memory key-value database without a @@ -34,9 +187,9 @@ func NewMemoryDatabase() ethdb.Database { return NewDatabase(memorydb.New()) } -// NewMemoryDatabaseWithCap creates an ephemeral in-memory key-value database with -// an initial starting capacity, but without a freezer moving immutable chain -// segments into cold storage. +// NewMemoryDatabaseWithCap creates an ephemeral in-memory key-value database +// with an initial starting capacity, but without a freezer moving immutable +// chain segments into cold storage. func NewMemoryDatabaseWithCap(size int) ethdb.Database { return NewDatabase(memorydb.NewWithCap(size)) } @@ -50,3 +203,153 @@ func NewLevelDBDatabase(file string, cache int, handles int, namespace string) ( } return NewDatabase(db), nil } + +// NewLevelDBDatabaseWithFreezer creates a persistent key-value database with a +// freezer moving immutable chain segments into cold storage. +func NewLevelDBDatabaseWithFreezer(file string, cache int, handles int, freezer string, namespace string) (ethdb.Database, error) { + kvdb, err := leveldb.New(file, cache, handles, namespace) + if err != nil { + return nil, err + } + frdb, err := NewDatabaseWithFreezer(kvdb, freezer, namespace) + if err != nil { + kvdb.Close() + return nil, err + } + return frdb, nil +} + +// InspectDatabase traverses the entire database and checks the size +// of all different categories of data. +func InspectDatabase(db ethdb.Database) error { + it := db.NewIterator() + defer it.Release() + + var ( + count int64 + start = time.Now() + logged = time.Now() + + // Key-value store statistics + total common.StorageSize + headerSize common.StorageSize + bodySize common.StorageSize + receiptSize common.StorageSize + tdSize common.StorageSize + numHashPairing common.StorageSize + hashNumPairing common.StorageSize + trieSize common.StorageSize + txlookupSize common.StorageSize + preimageSize common.StorageSize + bloomBitsSize common.StorageSize + cliqueSnapsSize common.StorageSize + + // Ancient store statistics + ancientHeaders common.StorageSize + ancientBodies common.StorageSize + ancientReceipts common.StorageSize + ancientHashes common.StorageSize + ancientTds common.StorageSize + + // Les statistic + chtTrieNodes common.StorageSize + bloomTrieNodes common.StorageSize + + // Meta- and unaccounted data + metadata common.StorageSize + unaccounted common.StorageSize + ) + // Inspect key-value database first. + for it.Next() { + var ( + key = it.Key() + size = common.StorageSize(len(key) + len(it.Value())) + ) + total += size + switch { + case bytes.HasPrefix(key, headerPrefix) && bytes.HasSuffix(key, headerTDSuffix): + tdSize += size + case bytes.HasPrefix(key, headerPrefix) && bytes.HasSuffix(key, headerHashSuffix): + numHashPairing += size + case bytes.HasPrefix(key, headerPrefix) && len(key) == (len(headerPrefix)+8+common.HashLength): + headerSize += size + case bytes.HasPrefix(key, headerNumberPrefix) && len(key) == (len(headerNumberPrefix)+common.HashLength): + hashNumPairing += size + case bytes.HasPrefix(key, blockBodyPrefix) && len(key) == (len(blockBodyPrefix)+8+common.HashLength): + bodySize += size + case bytes.HasPrefix(key, blockReceiptsPrefix) && len(key) == (len(blockReceiptsPrefix)+8+common.HashLength): + receiptSize += size + case bytes.HasPrefix(key, txLookupPrefix) && len(key) == (len(txLookupPrefix)+common.HashLength): + txlookupSize += size + case bytes.HasPrefix(key, preimagePrefix) && len(key) == (len(preimagePrefix)+common.HashLength): + preimageSize += size + case bytes.HasPrefix(key, bloomBitsPrefix) && len(key) == (len(bloomBitsPrefix)+10+common.HashLength): + bloomBitsSize += size + case bytes.HasPrefix(key, []byte("clique-")) && len(key) == 7+common.HashLength: + cliqueSnapsSize += size + case bytes.HasPrefix(key, []byte("cht-")) && len(key) == 4+common.HashLength: + chtTrieNodes += size + case bytes.HasPrefix(key, []byte("blt-")) && len(key) == 4+common.HashLength: + bloomTrieNodes += size + case len(key) == common.HashLength: + trieSize += size + default: + var accounted bool + for _, meta := range [][]byte{databaseVerisionKey, headHeaderKey, headBlockKey, headFastBlockKey, fastTrieProgressKey} { + if bytes.Equal(key, meta) { + metadata += size + accounted = true + break + } + } + if !accounted { + unaccounted += size + } + } + count += 1 + if count%1000 == 0 && time.Since(logged) > 8*time.Second { + log.Info("Inspecting database", "count", count, "elapsed", common.PrettyDuration(time.Since(start))) + logged = time.Now() + } + } + // Inspect append-only file store then. + ancients := []*common.StorageSize{&ancientHeaders, &ancientBodies, &ancientReceipts, &ancientHashes, &ancientTds} + for i, category := range []string{freezerHeaderTable, freezerBodiesTable, freezerReceiptTable, freezerHashTable, freezerDifficultyTable} { + if size, err := db.AncientSize(category); err == nil { + *ancients[i] += common.StorageSize(size) + total += common.StorageSize(size) + } + } + // Display the database statistic. + stats := [][]string{ + {"Key-Value store", "Headers", headerSize.String()}, + {"Key-Value store", "Bodies", bodySize.String()}, + {"Key-Value store", "Receipts", receiptSize.String()}, + {"Key-Value store", "Difficulties", tdSize.String()}, + {"Key-Value store", "Block number->hash", numHashPairing.String()}, + {"Key-Value store", "Block hash->number", hashNumPairing.String()}, + {"Key-Value store", "Transaction index", txlookupSize.String()}, + {"Key-Value store", "Bloombit index", bloomBitsSize.String()}, + {"Key-Value store", "Trie nodes", trieSize.String()}, + {"Key-Value store", "Trie preimages", preimageSize.String()}, + {"Key-Value store", "Clique snapshots", cliqueSnapsSize.String()}, + {"Key-Value store", "Singleton metadata", metadata.String()}, + {"Ancient store", "Headers", ancientHeaders.String()}, + {"Ancient store", "Bodies", ancientBodies.String()}, + {"Ancient store", "Receipts", ancientReceipts.String()}, + {"Ancient store", "Difficulties", ancientTds.String()}, + {"Ancient store", "Block number->hash", ancientHashes.String()}, + {"Light client", "CHT trie nodes", chtTrieNodes.String()}, + {"Light client", "Bloom trie nodes", bloomTrieNodes.String()}, + } + table := tablewriter.NewWriter(os.Stdout) + table.SetHeader([]string{"Database", "Category", "Size"}) + table.SetFooter([]string{"", "Total", total.String()}) + table.AppendBulk(stats) + table.Render() + + if unaccounted > 0 { + log.Error("Database contains unaccounted data", "size", unaccounted) + } + return nil +} diff --git a/core/rawdb/freezer.go b/core/rawdb/freezer.go new file mode 100644 index 000000000..67ed87d66 --- /dev/null +++ b/core/rawdb/freezer.go @@ -0,0 +1,382 @@ +// Copyright 2018 The go-ethereum Authors +// This file is part of the go-ethereum library. +// +// The go-ethereum library is free software: you can redistribute it and/or modify +// it under the terms of the GNU Lesser General Public License as published by +// the Free Software Foundation, either version 3 of the License, or +// (at your option) any later version. +// +// The go-ethereum library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public License +// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>. + +package rawdb + +import ( + "errors" + "fmt" + "math" + "os" + "path/filepath" + "sync/atomic" + "time" + + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/ethdb" + "github.com/ethereum/go-ethereum/log" + "github.com/ethereum/go-ethereum/metrics" + "github.com/ethereum/go-ethereum/params" + "github.com/prometheus/tsdb/fileutil" +) + +var ( + // errUnknownTable is returned if the user attempts to read from a table that is + // not tracked by the freezer. + errUnknownTable = errors.New("unknown table") + + // errOutOrderInsertion is returned if the user attempts to inject out-of-order + // binary blobs into the freezer. + errOutOrderInsertion = errors.New("the append operation is out-order") + + // errSymlinkDatadir is returned if the ancient directory specified by user + // is a symbolic link. + errSymlinkDatadir = errors.New("symbolic link datadir is not supported") +) + +const ( + // freezerRecheckInterval is the frequency to check the key-value database for + // chain progression that might permit new blocks to be frozen into immutable + // storage. + freezerRecheckInterval = time.Minute + + // freezerBatchLimit is the maximum number of blocks to freeze in one batch + // before doing an fsync and deleting it from the key-value store. + freezerBatchLimit = 30000 +) + +// freezer is an memory mapped append-only database to store immutable chain data +// into flat files: +// +// - The append only nature ensures that disk writes are minimized. +// - The memory mapping ensures we can max out system memory for caching without +// reserving it for go-ethereum. This would also reduce the memory requirements +// of Geth, and thus also GC overhead. +type freezer struct { + tables map[string]*freezerTable // Data tables for storing everything + frozen uint64 // Number of blocks already frozen + instanceLock fileutil.Releaser // File-system lock to prevent double opens +} + +// newFreezer creates a chain freezer that moves ancient chain data into +// append-only flat file containers. +func newFreezer(datadir string, namespace string) (*freezer, error) { + // Create the initial freezer object + var ( + readMeter = metrics.NewRegisteredMeter(namespace+"ancient/read", nil) + writeMeter = metrics.NewRegisteredMeter(namespace+"ancient/write", nil) + ) + // Ensure the datadir is not a symbolic link if it exists. + if info, err := os.Lstat(datadir); !os.IsNotExist(err) { + if info.Mode()&os.ModeSymlink != 0 { + log.Warn("Symbolic link ancient database is not supported", "path", datadir) + return nil, errSymlinkDatadir + } + } + // Leveldb uses LOCK as the filelock filename. To prevent the + // name collision, we use FLOCK as the lock name. + lock, _, err := fileutil.Flock(filepath.Join(datadir, "FLOCK")) + if err != nil { + return nil, err + } + // Open all the supported data tables + freezer := &freezer{ + tables: make(map[string]*freezerTable), + instanceLock: lock, + } + for name, disableSnappy := range freezerNoSnappy { + table, err := newTable(datadir, name, readMeter, writeMeter, disableSnappy) + if err != nil { + for _, table := range freezer.tables { + table.Close() + } + lock.Release() + return nil, err + } + freezer.tables[name] = table + } + if err := freezer.repair(); err != nil { + for _, table := range freezer.tables { + table.Close() + } + lock.Release() + return nil, err + } + log.Info("Opened ancient database", "database", datadir) + return freezer, nil +} + +// Close terminates the chain freezer, unmapping all the data files. +func (f *freezer) Close() error { + var errs []error + for _, table := range f.tables { + if err := table.Close(); err != nil { + errs = append(errs, err) + } + } + if err := f.instanceLock.Release(); err != nil { + errs = append(errs, err) + } + if errs != nil { + return fmt.Errorf("%v", errs) + } + return nil +} + +// HasAncient returns an indicator whether the specified ancient data exists +// in the freezer. +func (f *freezer) HasAncient(kind string, number uint64) (bool, error) { + if table := f.tables[kind]; table != nil { + return table.has(number), nil + } + return false, nil +} + +// Ancient retrieves an ancient binary blob from the append-only immutable files. +func (f *freezer) Ancient(kind string, number uint64) ([]byte, error) { + if table := f.tables[kind]; table != nil { + return table.Retrieve(number) + } + return nil, errUnknownTable +} + +// Ancients returns the length of the frozen items. +func (f *freezer) Ancients() (uint64, error) { + return atomic.LoadUint64(&f.frozen), nil +} + +// AncientSize returns the ancient size of the specified category. +func (f *freezer) AncientSize(kind string) (uint64, error) { + if table := f.tables[kind]; table != nil { + return table.size() + } + return 0, errUnknownTable +} + +// AppendAncient injects all binary blobs belong to block at the end of the +// append-only immutable table files. +// +// Notably, this function is lock free but kind of thread-safe. All out-of-order +// injection will be rejected. But if two injections with same number happen at +// the same time, we can get into the trouble. +func (f *freezer) AppendAncient(number uint64, hash, header, body, receipts, td []byte) (err error) { + // Ensure the binary blobs we are appending is continuous with freezer. + if atomic.LoadUint64(&f.frozen) != number { + return errOutOrderInsertion + } + // Rollback all inserted data if any insertion below failed to ensure + // the tables won't out of sync. + defer func() { + if err != nil { + rerr := f.repair() + if rerr != nil { + log.Crit("Failed to repair freezer", "err", rerr) + } + log.Info("Append ancient failed", "number", number, "err", err) + } + }() + // Inject all the components into the relevant data tables + if err := f.tables[freezerHashTable].Append(f.frozen, hash[:]); err != nil { + log.Error("Failed to append ancient hash", "number", f.frozen, "hash", hash, "err", err) + return err + } + if err := f.tables[freezerHeaderTable].Append(f.frozen, header); err != nil { + log.Error("Failed to append ancient header", "number", f.frozen, "hash", hash, "err", err) + return err + } + if err := f.tables[freezerBodiesTable].Append(f.frozen, body); err != nil { + log.Error("Failed to append ancient body", "number", f.frozen, "hash", hash, "err", err) + return err + } + if err := f.tables[freezerReceiptTable].Append(f.frozen, receipts); err != nil { + log.Error("Failed to append ancient receipts", "number", f.frozen, "hash", hash, "err", err) + return err + } + if err := f.tables[freezerDifficultyTable].Append(f.frozen, td); err != nil { + log.Error("Failed to append ancient difficulty", "number", f.frozen, "hash", hash, "err", err) + return err + } + atomic.AddUint64(&f.frozen, 1) // Only modify atomically + return nil +} + +// Truncate discards any recent data above the provided threshold number. +func (f *freezer) TruncateAncients(items uint64) error { + if atomic.LoadUint64(&f.frozen) <= items { + return nil + } + for _, table := range f.tables { + if err := table.truncate(items); err != nil { + return err + } + } + atomic.StoreUint64(&f.frozen, items) + return nil +} + +// sync flushes all data tables to disk. +func (f *freezer) Sync() error { + var errs []error + for _, table := range f.tables { + if err := table.Sync(); err != nil { + errs = append(errs, err) + } + } + if errs != nil { + return fmt.Errorf("%v", errs) + } + return nil +} + +// freeze is a background thread that periodically checks the blockchain for any +// import progress and moves ancient data from the fast database into the freezer. +// +// This functionality is deliberately broken off from block importing to avoid +// incurring additional data shuffling delays on block propagation. +func (f *freezer) freeze(db ethdb.KeyValueStore) { + nfdb := &nofreezedb{KeyValueStore: db} + + for { + // Retrieve the freezing threshold. + hash := ReadHeadBlockHash(nfdb) + if hash == (common.Hash{}) { + log.Debug("Current full block hash unavailable") // new chain, empty database + time.Sleep(freezerRecheckInterval) + continue + } + number := ReadHeaderNumber(nfdb, hash) + switch { + case number == nil: + log.Error("Current full block number unavailable", "hash", hash) + time.Sleep(freezerRecheckInterval) + continue + + case *number < params.ImmutabilityThreshold: + log.Debug("Current full block not old enough", "number", *number, "hash", hash, "delay", params.ImmutabilityThreshold) + time.Sleep(freezerRecheckInterval) + continue + + case *number-params.ImmutabilityThreshold <= f.frozen: + log.Debug("Ancient blocks frozen already", "number", *number, "hash", hash, "frozen", f.frozen) + time.Sleep(freezerRecheckInterval) + continue + } + head := ReadHeader(nfdb, hash, *number) + if head == nil { + log.Error("Current full block unavailable", "number", *number, "hash", hash) + time.Sleep(freezerRecheckInterval) + continue + } + // Seems we have data ready to be frozen, process in usable batches + limit := *number - params.ImmutabilityThreshold + if limit-f.frozen > freezerBatchLimit { + limit = f.frozen + freezerBatchLimit + } + var ( + start = time.Now() + first = f.frozen + ancients = make([]common.Hash, 0, limit) + ) + for f.frozen < limit { + // Retrieves all the components of the canonical block + hash := ReadCanonicalHash(nfdb, f.frozen) + if hash == (common.Hash{}) { + log.Error("Canonical hash missing, can't freeze", "number", f.frozen) + break + } + header := ReadHeaderRLP(nfdb, hash, f.frozen) + if len(header) == 0 { + log.Error("Block header missing, can't freeze", "number", f.frozen, "hash", hash) + break + } + body := ReadBodyRLP(nfdb, hash, f.frozen) + if len(body) == 0 { + log.Error("Block body missing, can't freeze", "number", f.frozen, "hash", hash) + break + } + receipts := ReadReceiptsRLP(nfdb, hash, f.frozen) + if len(receipts) == 0 { + log.Error("Block receipts missing, can't freeze", "number", f.frozen, "hash", hash) + break + } + td := ReadTdRLP(nfdb, hash, f.frozen) + if len(td) == 0 { + log.Error("Total difficulty missing, can't freeze", "number", f.frozen, "hash", hash) + break + } + log.Trace("Deep froze ancient block", "number", f.frozen, "hash", hash) + // Inject all the components into the relevant data tables + if err := f.AppendAncient(f.frozen, hash[:], header, body, receipts, td); err != nil { + break + } + ancients = append(ancients, hash) + } + // Batch of blocks have been frozen, flush them before wiping from leveldb + if err := f.Sync(); err != nil { + log.Crit("Failed to flush frozen tables", "err", err) + } + // Wipe out all data from the active database + batch := db.NewBatch() + for i := 0; i < len(ancients); i++ { + DeleteBlockWithoutNumber(batch, ancients[i], first+uint64(i)) + DeleteCanonicalHash(batch, first+uint64(i)) + } + if err := batch.Write(); err != nil { + log.Crit("Failed to delete frozen canonical blocks", "err", err) + } + batch.Reset() + // Wipe out side chain also. + for number := first; number < f.frozen; number++ { + for _, hash := range ReadAllHashes(db, number) { + DeleteBlock(batch, hash, number) + } + } + if err := batch.Write(); err != nil { + log.Crit("Failed to delete frozen side blocks", "err", err) + } + // Log something friendly for the user + context := []interface{}{ + "blocks", f.frozen - first, "elapsed", common.PrettyDuration(time.Since(start)), "number", f.frozen - 1, + } + if n := len(ancients); n > 0 { + context = append(context, []interface{}{"hash", ancients[n-1]}...) + } + log.Info("Deep froze chain segment", context...) + + // Avoid database thrashing with tiny writes + if f.frozen-first < freezerBatchLimit { + time.Sleep(freezerRecheckInterval) + } + } +} + +// repair truncates all data tables to the same length. +func (f *freezer) repair() error { + min := uint64(math.MaxUint64) + for _, table := range f.tables { + items := atomic.LoadUint64(&table.items) + if min > items { + min = items + } + } + for _, table := range f.tables { + if err := table.truncate(min); err != nil { + return err + } + } + atomic.StoreUint64(&f.frozen, min) + return nil +} diff --git a/core/rawdb/freezer_table.go b/core/rawdb/freezer_table.go new file mode 100644 index 000000000..673a181e4 --- /dev/null +++ b/core/rawdb/freezer_table.go @@ -0,0 +1,561 @@ +// Copyright 2019 The go-ethereum Authors +// This file is part of the go-ethereum library. +// +// The go-ethereum library is free software: you can redistribute it and/or modify +// it under the terms of the GNU Lesser General Public License as published by +// the Free Software Foundation, either version 3 of the License, or +// (at your option) any later version. +// +// The go-ethereum library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public License +// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>. + +package rawdb + +import ( + "encoding/binary" + "errors" + "fmt" + "os" + "path/filepath" + "sync" + "sync/atomic" + + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/log" + "github.com/ethereum/go-ethereum/metrics" + "github.com/golang/snappy" +) + +var ( + // errClosed is returned if an operation attempts to read from or write to the + // freezer table after it has already been closed. + errClosed = errors.New("closed") + + // errOutOfBounds is returned if the item requested is not contained within the + // freezer table. + errOutOfBounds = errors.New("out of bounds") + + // errNotSupported is returned if the database doesn't support the required operation. + errNotSupported = errors.New("this operation is not supported") +) + +// indexEntry contains the number/id of the file that the data resides in, aswell as the +// offset within the file to the end of the data +// In serialized form, the filenum is stored as uint16. +type indexEntry struct { + filenum uint32 // stored as uint16 ( 2 bytes) + offset uint32 // stored as uint32 ( 4 bytes) +} + +const indexEntrySize = 6 + +// unmarshallBinary deserializes binary b into the rawIndex entry. +func (i *indexEntry) unmarshalBinary(b []byte) error { + i.filenum = uint32(binary.BigEndian.Uint16(b[:2])) + i.offset = binary.BigEndian.Uint32(b[2:6]) + return nil +} + +// marshallBinary serializes the rawIndex entry into binary. +func (i *indexEntry) marshallBinary() []byte { + b := make([]byte, indexEntrySize) + binary.BigEndian.PutUint16(b[:2], uint16(i.filenum)) + binary.BigEndian.PutUint32(b[2:6], i.offset) + return b +} + +// freezerTable represents a single chained data table within the freezer (e.g. blocks). +// It consists of a data file (snappy encoded arbitrary data blobs) and an indexEntry +// file (uncompressed 64 bit indices into the data file). +type freezerTable struct { + noCompression bool // if true, disables snappy compression. Note: does not work retroactively + maxFileSize uint32 // Max file size for data-files + name string + path string + + head *os.File // File descriptor for the data head of the table + files map[uint32]*os.File // open files + headId uint32 // number of the currently active head file + tailId uint32 // number of the earliest file + index *os.File // File descriptor for the indexEntry file of the table + + // In the case that old items are deleted (from the tail), we use itemOffset + // to count how many historic items have gone missing. + items uint64 // Number of items stored in the table (including items removed from tail) + itemOffset uint32 // Offset (number of discarded items) + + headBytes uint32 // Number of bytes written to the head file + readMeter metrics.Meter // Meter for measuring the effective amount of data read + writeMeter metrics.Meter // Meter for measuring the effective amount of data written + + logger log.Logger // Logger with database path and table name ambedded + lock sync.RWMutex // Mutex protecting the data file descriptors +} + +// newTable opens a freezer table with default settings - 2G files +func newTable(path string, name string, readMeter metrics.Meter, writeMeter metrics.Meter, disableSnappy bool) (*freezerTable, error) { + return newCustomTable(path, name, readMeter, writeMeter, 2*1000*1000*1000, disableSnappy) +} + +// newCustomTable opens a freezer table, creating the data and index files if they are +// non existent. Both files are truncated to the shortest common length to ensure +// they don't go out of sync. +func newCustomTable(path string, name string, readMeter metrics.Meter, writeMeter metrics.Meter, maxFilesize uint32, noCompression bool) (*freezerTable, error) { + // Ensure the containing directory exists and open the indexEntry file + if err := os.MkdirAll(path, 0755); err != nil { + return nil, err + } + var idxName string + if noCompression { + // raw idx + idxName = fmt.Sprintf("%s.ridx", name) + } else { + // compressed idx + idxName = fmt.Sprintf("%s.cidx", name) + } + offsets, err := os.OpenFile(filepath.Join(path, idxName), os.O_RDWR|os.O_CREATE|os.O_APPEND, 0644) + if err != nil { + return nil, err + } + // Create the table and repair any past inconsistency + tab := &freezerTable{ + index: offsets, + files: make(map[uint32]*os.File), + readMeter: readMeter, + writeMeter: writeMeter, + name: name, + path: path, + logger: log.New("database", path, "table", name), + noCompression: noCompression, + maxFileSize: maxFilesize, + } + if err := tab.repair(); err != nil { + tab.Close() + return nil, err + } + return tab, nil +} + +// repair cross checks the head and the index file and truncates them to +// be in sync with each other after a potential crash / data loss. +func (t *freezerTable) repair() error { + // Create a temporary offset buffer to init files with and read indexEntry into + buffer := make([]byte, indexEntrySize) + + // If we've just created the files, initialize the index with the 0 indexEntry + stat, err := t.index.Stat() + if err != nil { + return err + } + if stat.Size() == 0 { + if _, err := t.index.Write(buffer); err != nil { + return err + } + } + // Ensure the index is a multiple of indexEntrySize bytes + if overflow := stat.Size() % indexEntrySize; overflow != 0 { + t.index.Truncate(stat.Size() - overflow) // New file can't trigger this path + } + // Retrieve the file sizes and prepare for truncation + if stat, err = t.index.Stat(); err != nil { + return err + } + offsetsSize := stat.Size() + + // Open the head file + var ( + firstIndex indexEntry + lastIndex indexEntry + contentSize int64 + contentExp int64 + ) + // Read index zero, determine what file is the earliest + // and what item offset to use + t.index.ReadAt(buffer, 0) + firstIndex.unmarshalBinary(buffer) + + t.tailId = firstIndex.offset + t.itemOffset = firstIndex.filenum + + t.index.ReadAt(buffer, offsetsSize-indexEntrySize) + lastIndex.unmarshalBinary(buffer) + t.head, err = t.openFile(lastIndex.filenum, os.O_RDWR|os.O_CREATE|os.O_APPEND) + if err != nil { + return err + } + if stat, err = t.head.Stat(); err != nil { + return err + } + contentSize = stat.Size() + + // Keep truncating both files until they come in sync + contentExp = int64(lastIndex.offset) + + for contentExp != contentSize { + // Truncate the head file to the last offset pointer + if contentExp < contentSize { + t.logger.Warn("Truncating dangling head", "indexed", common.StorageSize(contentExp), "stored", common.StorageSize(contentSize)) + if err := t.head.Truncate(contentExp); err != nil { + return err + } + contentSize = contentExp + } + // Truncate the index to point within the head file + if contentExp > contentSize { + t.logger.Warn("Truncating dangling indexes", "indexed", common.StorageSize(contentExp), "stored", common.StorageSize(contentSize)) + if err := t.index.Truncate(offsetsSize - indexEntrySize); err != nil { + return err + } + offsetsSize -= indexEntrySize + t.index.ReadAt(buffer, offsetsSize-indexEntrySize) + var newLastIndex indexEntry + newLastIndex.unmarshalBinary(buffer) + // We might have slipped back into an earlier head-file here + if newLastIndex.filenum != lastIndex.filenum { + // release earlier opened file + t.releaseFile(lastIndex.filenum) + t.head, err = t.openFile(newLastIndex.filenum, os.O_RDWR|os.O_CREATE|os.O_APPEND) + if stat, err = t.head.Stat(); err != nil { + // TODO, anything more we can do here? + // A data file has gone missing... + return err + } + contentSize = stat.Size() + } + lastIndex = newLastIndex + contentExp = int64(lastIndex.offset) + } + } + // Ensure all reparation changes have been written to disk + if err := t.index.Sync(); err != nil { + return err + } + if err := t.head.Sync(); err != nil { + return err + } + // Update the item and byte counters and return + t.items = uint64(t.itemOffset) + uint64(offsetsSize/indexEntrySize-1) // last indexEntry points to the end of the data file + t.headBytes = uint32(contentSize) + t.headId = lastIndex.filenum + + // Close opened files and preopen all files + if err := t.preopen(); err != nil { + return err + } + t.logger.Debug("Chain freezer table opened", "items", t.items, "size", common.StorageSize(t.headBytes)) + return nil +} + +// preopen opens all files that the freezer will need. This method should be called from an init-context, +// since it assumes that it doesn't have to bother with locking +// The rationale for doing preopen is to not have to do it from within Retrieve, thus not needing to ever +// obtain a write-lock within Retrieve. +func (t *freezerTable) preopen() (err error) { + // The repair might have already opened (some) files + t.releaseFilesAfter(0, false) + // Open all except head in RDONLY + for i := t.tailId; i < t.headId; i++ { + if _, err = t.openFile(i, os.O_RDONLY); err != nil { + return err + } + } + // Open head in read/write + t.head, err = t.openFile(t.headId, os.O_RDWR|os.O_CREATE|os.O_APPEND) + return err +} + +// truncate discards any recent data above the provided threashold number. +func (t *freezerTable) truncate(items uint64) error { + t.lock.Lock() + defer t.lock.Unlock() + + // If our item count is correct, don't do anything + if atomic.LoadUint64(&t.items) <= items { + return nil + } + // Something's out of sync, truncate the table's offset index + t.logger.Warn("Truncating freezer table", "items", t.items, "limit", items) + if err := t.index.Truncate(int64(items+1) * indexEntrySize); err != nil { + return err + } + // Calculate the new expected size of the data file and truncate it + buffer := make([]byte, indexEntrySize) + if _, err := t.index.ReadAt(buffer, int64(items*indexEntrySize)); err != nil { + return err + } + var expected indexEntry + expected.unmarshalBinary(buffer) + + // We might need to truncate back to older files + if expected.filenum != t.headId { + // If already open for reading, force-reopen for writing + t.releaseFile(expected.filenum) + newHead, err := t.openFile(expected.filenum, os.O_RDWR|os.O_CREATE|os.O_APPEND) + if err != nil { + return err + } + // release any files _after the current head -- both the previous head + // and any files which may have been opened for reading + t.releaseFilesAfter(expected.filenum, true) + // set back the historic head + t.head = newHead + atomic.StoreUint32(&t.headId, expected.filenum) + } + if err := t.head.Truncate(int64(expected.offset)); err != nil { + return err + } + // All data files truncated, set internal counters and return + atomic.StoreUint64(&t.items, items) + atomic.StoreUint32(&t.headBytes, expected.offset) + return nil +} + +// Close closes all opened files. +func (t *freezerTable) Close() error { + t.lock.Lock() + defer t.lock.Unlock() + + var errs []error + if err := t.index.Close(); err != nil { + errs = append(errs, err) + } + t.index = nil + + for _, f := range t.files { + if err := f.Close(); err != nil { + errs = append(errs, err) + } + } + t.head = nil + + if errs != nil { + return fmt.Errorf("%v", errs) + } + return nil +} + +// openFile assumes that the write-lock is held by the caller +func (t *freezerTable) openFile(num uint32, flag int) (f *os.File, err error) { + var exist bool + if f, exist = t.files[num]; !exist { + var name string + if t.noCompression { + name = fmt.Sprintf("%s.%04d.rdat", t.name, num) + } else { + name = fmt.Sprintf("%s.%04d.cdat", t.name, num) + } + f, err = os.OpenFile(filepath.Join(t.path, name), flag, 0644) + if err != nil { + return nil, err + } + t.files[num] = f + } + return f, err +} + +// releaseFile closes a file, and removes it from the open file cache. +// Assumes that the caller holds the write lock +func (t *freezerTable) releaseFile(num uint32) { + if f, exist := t.files[num]; exist { + delete(t.files, num) + f.Close() + } +} + +// releaseFilesAfter closes all open files with a higher number, and optionally also deletes the files +func (t *freezerTable) releaseFilesAfter(num uint32, remove bool) { + for fnum, f := range t.files { + if fnum > num { + delete(t.files, fnum) + f.Close() + if remove { + os.Remove(f.Name()) + } + } + } +} + +// Append injects a binary blob at the end of the freezer table. The item number +// is a precautionary parameter to ensure data correctness, but the table will +// reject already existing data. +// +// Note, this method will *not* flush any data to disk so be sure to explicitly +// fsync before irreversibly deleting data from the database. +func (t *freezerTable) Append(item uint64, blob []byte) error { + // Read lock prevents competition with truncate + t.lock.RLock() + // Ensure the table is still accessible + if t.index == nil || t.head == nil { + t.lock.RUnlock() + return errClosed + } + // Ensure only the next item can be written, nothing else + if atomic.LoadUint64(&t.items) != item { + t.lock.RUnlock() + return fmt.Errorf("appending unexpected item: want %d, have %d", t.items, item) + } + // Encode the blob and write it into the data file + if !t.noCompression { + blob = snappy.Encode(nil, blob) + } + bLen := uint32(len(blob)) + if t.headBytes+bLen < bLen || + t.headBytes+bLen > t.maxFileSize { + // we need a new file, writing would overflow + t.lock.RUnlock() + t.lock.Lock() + nextId := atomic.LoadUint32(&t.headId) + 1 + // We open the next file in truncated mode -- if this file already + // exists, we need to start over from scratch on it + newHead, err := t.openFile(nextId, os.O_RDWR|os.O_CREATE|os.O_TRUNC) + if err != nil { + t.lock.Unlock() + return err + } + // Close old file, and reopen in RDONLY mode + t.releaseFile(t.headId) + t.openFile(t.headId, os.O_RDONLY) + + // Swap out the current head + t.head = newHead + atomic.StoreUint32(&t.headBytes, 0) + atomic.StoreUint32(&t.headId, nextId) + t.lock.Unlock() + t.lock.RLock() + } + + defer t.lock.RUnlock() + + if _, err := t.head.Write(blob); err != nil { + return err + } + newOffset := atomic.AddUint32(&t.headBytes, bLen) + idx := indexEntry{ + filenum: atomic.LoadUint32(&t.headId), + offset: newOffset, + } + // Write indexEntry + t.index.Write(idx.marshallBinary()) + t.writeMeter.Mark(int64(bLen + indexEntrySize)) + atomic.AddUint64(&t.items, 1) + return nil +} + +// getBounds returns the indexes for the item +// returns start, end, filenumber and error +func (t *freezerTable) getBounds(item uint64) (uint32, uint32, uint32, error) { + var startIdx, endIdx indexEntry + buffer := make([]byte, indexEntrySize) + if _, err := t.index.ReadAt(buffer, int64(item*indexEntrySize)); err != nil { + return 0, 0, 0, err + } + startIdx.unmarshalBinary(buffer) + if _, err := t.index.ReadAt(buffer, int64((item+1)*indexEntrySize)); err != nil { + return 0, 0, 0, err + } + endIdx.unmarshalBinary(buffer) + if startIdx.filenum != endIdx.filenum { + // If a piece of data 'crosses' a data-file, + // it's actually in one piece on the second data-file. + // We return a zero-indexEntry for the second file as start + return 0, endIdx.offset, endIdx.filenum, nil + } + return startIdx.offset, endIdx.offset, endIdx.filenum, nil +} + +// Retrieve looks up the data offset of an item with the given number and retrieves +// the raw binary blob from the data file. +func (t *freezerTable) Retrieve(item uint64) ([]byte, error) { + // Ensure the table and the item is accessible + if t.index == nil || t.head == nil { + return nil, errClosed + } + if atomic.LoadUint64(&t.items) <= item { + return nil, errOutOfBounds + } + // Ensure the item was not deleted from the tail either + offset := atomic.LoadUint32(&t.itemOffset) + if uint64(offset) > item { + return nil, errOutOfBounds + } + t.lock.RLock() + startOffset, endOffset, filenum, err := t.getBounds(item - uint64(offset)) + if err != nil { + t.lock.RUnlock() + return nil, err + } + dataFile, exist := t.files[filenum] + if !exist { + t.lock.RUnlock() + return nil, fmt.Errorf("missing data file %d", filenum) + } + // Retrieve the data itself, decompress and return + blob := make([]byte, endOffset-startOffset) + if _, err := dataFile.ReadAt(blob, int64(startOffset)); err != nil { + t.lock.RUnlock() + return nil, err + } + t.lock.RUnlock() + t.readMeter.Mark(int64(len(blob) + 2*indexEntrySize)) + + if t.noCompression { + return blob, nil + } + return snappy.Decode(nil, blob) +} + +// has returns an indicator whether the specified number data +// exists in the freezer table. +func (t *freezerTable) has(number uint64) bool { + return atomic.LoadUint64(&t.items) > number +} + +// size returns the total data size in the freezer table. +func (t *freezerTable) size() (uint64, error) { + t.lock.RLock() + defer t.lock.RUnlock() + + stat, err := t.index.Stat() + if err != nil { + return 0, err + } + total := uint64(t.maxFileSize)*uint64(t.headId-t.tailId) + uint64(t.headBytes) + uint64(stat.Size()) + return total, nil +} + +// Sync pushes any pending data from memory out to disk. This is an expensive +// operation, so use it with care. +func (t *freezerTable) Sync() error { + if err := t.index.Sync(); err != nil { + return err + } + return t.head.Sync() +} + +// printIndex is a debug print utility function for testing +func (t *freezerTable) printIndex() { + buf := make([]byte, indexEntrySize) + + fmt.Printf("|-----------------|\n") + fmt.Printf("| fileno | offset |\n") + fmt.Printf("|--------+--------|\n") + + for i := uint64(0); ; i++ { + if _, err := t.index.ReadAt(buf, int64(i*indexEntrySize)); err != nil { + break + } + var entry indexEntry + entry.unmarshalBinary(buf) + fmt.Printf("| %03d | %03d | \n", entry.filenum, entry.offset) + if i > 100 { + fmt.Printf(" ... \n") + break + } + } + fmt.Printf("|-----------------|\n") +} diff --git a/core/rawdb/freezer_table_test.go b/core/rawdb/freezer_table_test.go new file mode 100644 index 000000000..e63fb63a3 --- /dev/null +++ b/core/rawdb/freezer_table_test.go @@ -0,0 +1,609 @@ +// Copyright 2018 The go-ethereum Authors +// This file is part of the go-ethereum library. +// +// The go-ethereum library is free software: you can redistribute it and/or modify +// it under the terms of the GNU Lesser General Public License as published by +// the Free Software Foundation, either version 3 of the License, or +// (at your option) any later version. +// +// The go-ethereum library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public License +// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>. + +package rawdb + +import ( + "bytes" + "fmt" + "math/rand" + "os" + "path/filepath" + "testing" + "time" + + "github.com/ethereum/go-ethereum/metrics" +) + +func init() { + rand.Seed(time.Now().Unix()) +} + +// Gets a chunk of data, filled with 'b' +func getChunk(size int, b int) []byte { + data := make([]byte, size) + for i := range data { + data[i] = byte(b) + } + return data +} + +func print(t *testing.T, f *freezerTable, item uint64) { + a, err := f.Retrieve(item) + if err != nil { + t.Fatal(err) + } + fmt.Printf("db[%d] = %x\n", item, a) +} + +// TestFreezerBasics test initializing a freezertable from scratch, writing to the table, +// and reading it back. +func TestFreezerBasics(t *testing.T) { + t.Parallel() + // set cutoff at 50 bytes + f, err := newCustomTable(os.TempDir(), + fmt.Sprintf("unittest-%d", rand.Uint64()), + metrics.NewMeter(), metrics.NewMeter(), 50, true) + if err != nil { + t.Fatal(err) + } + defer f.Close() + // Write 15 bytes 255 times, results in 85 files + for x := 0; x < 255; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + } + + //print(t, f, 0) + //print(t, f, 1) + //print(t, f, 2) + // + //db[0] = 000000000000000000000000000000 + //db[1] = 010101010101010101010101010101 + //db[2] = 020202020202020202020202020202 + + for y := 0; y < 255; y++ { + exp := getChunk(15, y) + got, err := f.Retrieve(uint64(y)) + if err != nil { + t.Fatal(err) + } + if !bytes.Equal(got, exp) { + t.Fatalf("test %d, got \n%x != \n%x", y, got, exp) + } + } + // Check that we cannot read too far + _, err = f.Retrieve(uint64(255)) + if err != errOutOfBounds { + t.Fatal(err) + } +} + +// TestFreezerBasicsClosing tests same as TestFreezerBasics, but also closes and reopens the freezer between +// every operation +func TestFreezerBasicsClosing(t *testing.T) { + t.Parallel() + // set cutoff at 50 bytes + var ( + fname = fmt.Sprintf("basics-close-%d", rand.Uint64()) + m1, m2 = metrics.NewMeter(), metrics.NewMeter() + f *freezerTable + err error + ) + f, err = newCustomTable(os.TempDir(), fname, m1, m2, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 15 bytes 255 times, results in 85 files + for x := 0; x < 255; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + f.Close() + f, err = newCustomTable(os.TempDir(), fname, m1, m2, 50, true) + } + defer f.Close() + + for y := 0; y < 255; y++ { + exp := getChunk(15, y) + got, err := f.Retrieve(uint64(y)) + if err != nil { + t.Fatal(err) + } + if !bytes.Equal(got, exp) { + t.Fatalf("test %d, got \n%x != \n%x", y, got, exp) + } + f.Close() + f, err = newCustomTable(os.TempDir(), fname, m1, m2, 50, true) + if err != nil { + t.Fatal(err) + } + } +} + +// TestFreezerRepairDanglingHead tests that we can recover if index entries are removed +func TestFreezerRepairDanglingHead(t *testing.T) { + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("dangling_headtest-%d", rand.Uint64()) + + { // Fill table + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 15 bytes 255 times + for x := 0; x < 255; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + } + // The last item should be there + if _, err = f.Retrieve(0xfe); err != nil { + t.Fatal(err) + } + f.Close() + } + // open the index + idxFile, err := os.OpenFile(filepath.Join(os.TempDir(), fmt.Sprintf("%s.ridx", fname)), os.O_RDWR, 0644) + if err != nil { + t.Fatalf("Failed to open index file: %v", err) + } + // Remove 4 bytes + stat, err := idxFile.Stat() + if err != nil { + t.Fatalf("Failed to stat index file: %v", err) + } + idxFile.Truncate(stat.Size() - 4) + idxFile.Close() + // Now open it again + { + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + // The last item should be missing + if _, err = f.Retrieve(0xff); err == nil { + t.Errorf("Expected error for missing index entry") + } + // The one before should still be there + if _, err = f.Retrieve(0xfd); err != nil { + t.Fatalf("Expected no error, got %v", err) + } + } +} + +// TestFreezerRepairDanglingHeadLarge tests that we can recover if very many index entries are removed +func TestFreezerRepairDanglingHeadLarge(t *testing.T) { + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("dangling_headtest-%d", rand.Uint64()) + + { // Fill a table and close it + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 15 bytes 255 times + for x := 0; x < 0xff; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + } + // The last item should be there + if _, err = f.Retrieve(f.items - 1); err == nil { + if err != nil { + t.Fatal(err) + } + } + f.Close() + } + // open the index + idxFile, err := os.OpenFile(filepath.Join(os.TempDir(), fmt.Sprintf("%s.ridx", fname)), os.O_RDWR, 0644) + if err != nil { + t.Fatalf("Failed to open index file: %v", err) + } + // Remove everything but the first item, and leave data unaligned + // 0-indexEntry, 1-indexEntry, corrupt-indexEntry + idxFile.Truncate(indexEntrySize + indexEntrySize + indexEntrySize/2) + idxFile.Close() + // Now open it again + { + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + // The first item should be there + if _, err = f.Retrieve(0); err != nil { + t.Fatal(err) + } + // The second item should be missing + if _, err = f.Retrieve(1); err == nil { + t.Errorf("Expected error for missing index entry") + } + // We should now be able to store items again, from item = 1 + for x := 1; x < 0xff; x++ { + data := getChunk(15, ^x) + f.Append(uint64(x), data) + } + f.Close() + } + // And if we open it, we should now be able to read all of them (new values) + { + f, _ := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + for y := 1; y < 255; y++ { + exp := getChunk(15, ^y) + got, err := f.Retrieve(uint64(y)) + if err != nil { + t.Fatal(err) + } + if !bytes.Equal(got, exp) { + t.Fatalf("test %d, got \n%x != \n%x", y, got, exp) + } + } + } +} + +// TestSnappyDetection tests that we fail to open a snappy database and vice versa +func TestSnappyDetection(t *testing.T) { + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("snappytest-%d", rand.Uint64()) + // Open with snappy + { + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 15 bytes 255 times + for x := 0; x < 0xff; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + } + f.Close() + } + // Open without snappy + { + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, false) + if _, err = f.Retrieve(0); err == nil { + f.Close() + t.Fatalf("expected empty table") + } + } + + // Open with snappy + { + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, true) + // There should be 255 items + if _, err = f.Retrieve(0xfe); err != nil { + f.Close() + t.Fatalf("expected no error, got %v", err) + } + } + +} +func assertFileSize(f string, size int64) error { + stat, err := os.Stat(f) + if err != nil { + return err + } + if stat.Size() != size { + return fmt.Errorf("error, expected size %d, got %d", size, stat.Size()) + } + return nil + +} + +// TestFreezerRepairDanglingIndex checks that if the index has more entries than there are data, +// the index is repaired +func TestFreezerRepairDanglingIndex(t *testing.T) { + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("dangling_indextest-%d", rand.Uint64()) + + { // Fill a table and close it + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 15 bytes 9 times : 150 bytes + for x := 0; x < 9; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + } + // The last item should be there + if _, err = f.Retrieve(f.items - 1); err != nil { + f.Close() + t.Fatal(err) + } + f.Close() + // File sizes should be 45, 45, 45 : items[3, 3, 3) + } + // Crop third file + fileToCrop := filepath.Join(os.TempDir(), fmt.Sprintf("%s.0002.rdat", fname)) + // Truncate third file: 45 ,45, 20 + { + if err := assertFileSize(fileToCrop, 45); err != nil { + t.Fatal(err) + } + file, err := os.OpenFile(fileToCrop, os.O_RDWR, 0644) + if err != nil { + t.Fatal(err) + } + file.Truncate(20) + file.Close() + } + // Open db it again + // It should restore the file(s) to + // 45, 45, 15 + // with 3+3+1 items + { + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, true) + if err != nil { + t.Fatal(err) + } + if f.items != 7 { + f.Close() + t.Fatalf("expected %d items, got %d", 7, f.items) + } + if err := assertFileSize(fileToCrop, 15); err != nil { + t.Fatal(err) + } + } +} + +func TestFreezerTruncate(t *testing.T) { + + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("truncation-%d", rand.Uint64()) + + { // Fill table + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 15 bytes 30 times + for x := 0; x < 30; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + } + // The last item should be there + if _, err = f.Retrieve(f.items - 1); err != nil { + t.Fatal(err) + } + f.Close() + } + // Reopen, truncate + { + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + if err != nil { + t.Fatal(err) + } + defer f.Close() + f.truncate(10) // 150 bytes + if f.items != 10 { + t.Fatalf("expected %d items, got %d", 10, f.items) + } + // 45, 45, 45, 15 -- bytes should be 15 + if f.headBytes != 15 { + t.Fatalf("expected %d bytes, got %d", 15, f.headBytes) + } + + } + +} + +// TestFreezerRepairFirstFile tests a head file with the very first item only half-written. +// That will rewind the index, and _should_ truncate the head file +func TestFreezerRepairFirstFile(t *testing.T) { + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("truncationfirst-%d", rand.Uint64()) + { // Fill table + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 80 bytes, splitting out into two files + f.Append(0, getChunk(40, 0xFF)) + f.Append(1, getChunk(40, 0xEE)) + // The last item should be there + if _, err = f.Retrieve(f.items - 1); err != nil { + t.Fatal(err) + } + f.Close() + } + // Truncate the file in half + fileToCrop := filepath.Join(os.TempDir(), fmt.Sprintf("%s.0001.rdat", fname)) + { + if err := assertFileSize(fileToCrop, 40); err != nil { + t.Fatal(err) + } + file, err := os.OpenFile(fileToCrop, os.O_RDWR, 0644) + if err != nil { + t.Fatal(err) + } + file.Truncate(20) + file.Close() + } + // Reopen + { + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, true) + if err != nil { + t.Fatal(err) + } + if f.items != 1 { + f.Close() + t.Fatalf("expected %d items, got %d", 0, f.items) + } + // Write 40 bytes + f.Append(1, getChunk(40, 0xDD)) + f.Close() + // Should have been truncated down to zero and then 40 written + if err := assertFileSize(fileToCrop, 40); err != nil { + t.Fatal(err) + } + } +} + +// TestFreezerReadAndTruncate tests: +// - we have a table open +// - do some reads, so files are open in readonly +// - truncate so those files are 'removed' +// - check that we did not keep the rdonly file descriptors +func TestFreezerReadAndTruncate(t *testing.T) { + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("read_truncate-%d", rand.Uint64()) + { // Fill table + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 50, true) + if err != nil { + t.Fatal(err) + } + // Write 15 bytes 30 times + for x := 0; x < 30; x++ { + data := getChunk(15, x) + f.Append(uint64(x), data) + } + // The last item should be there + if _, err = f.Retrieve(f.items - 1); err != nil { + t.Fatal(err) + } + f.Close() + } + // Reopen and read all files + { + f, err := newCustomTable(os.TempDir(), fname, wm, rm, 50, true) + if err != nil { + t.Fatal(err) + } + if f.items != 30 { + f.Close() + t.Fatalf("expected %d items, got %d", 0, f.items) + } + for y := byte(0); y < 30; y++ { + f.Retrieve(uint64(y)) + } + // Now, truncate back to zero + f.truncate(0) + // Write the data again + for x := 0; x < 30; x++ { + data := getChunk(15, ^x) + if err := f.Append(uint64(x), data); err != nil { + t.Fatalf("error %v", err) + } + } + f.Close() + } +} + +func TestOffset(t *testing.T) { + t.Parallel() + wm, rm := metrics.NewMeter(), metrics.NewMeter() + fname := fmt.Sprintf("offset-%d", rand.Uint64()) + { // Fill table + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 40, true) + if err != nil { + t.Fatal(err) + } + // Write 6 x 20 bytes, splitting out into three files + f.Append(0, getChunk(20, 0xFF)) + f.Append(1, getChunk(20, 0xEE)) + + f.Append(2, getChunk(20, 0xdd)) + f.Append(3, getChunk(20, 0xcc)) + + f.Append(4, getChunk(20, 0xbb)) + f.Append(5, getChunk(20, 0xaa)) + f.printIndex() + f.Close() + } + // Now crop it. + { + // delete files 0 and 1 + for i := 0; i < 2; i++ { + p := filepath.Join(os.TempDir(), fmt.Sprintf("%v.%04d.rdat", fname, i)) + if err := os.Remove(p); err != nil { + t.Fatal(err) + } + } + // Read the index file + p := filepath.Join(os.TempDir(), fmt.Sprintf("%v.ridx", fname)) + indexFile, err := os.OpenFile(p, os.O_RDWR, 0644) + if err != nil { + t.Fatal(err) + } + indexBuf := make([]byte, 7*indexEntrySize) + indexFile.Read(indexBuf) + + // Update the index file, so that we store + // [ file = 2, offset = 4 ] at index zero + + tailId := uint32(2) // First file is 2 + itemOffset := uint32(4) // We have removed four items + zeroIndex := indexEntry{ + offset: tailId, + filenum: itemOffset, + } + buf := zeroIndex.marshallBinary() + // Overwrite index zero + copy(indexBuf, buf) + // Remove the four next indices by overwriting + copy(indexBuf[indexEntrySize:], indexBuf[indexEntrySize*5:]) + indexFile.WriteAt(indexBuf, 0) + // Need to truncate the moved index items + indexFile.Truncate(indexEntrySize * (1 + 2)) + indexFile.Close() + + } + // Now open again + { + f, err := newCustomTable(os.TempDir(), fname, rm, wm, 40, true) + if err != nil { + t.Fatal(err) + } + f.printIndex() + // It should allow writing item 6 + f.Append(6, getChunk(20, 0x99)) + + // It should be fine to fetch 4,5,6 + if got, err := f.Retrieve(4); err != nil { + t.Fatal(err) + } else if exp := getChunk(20, 0xbb); !bytes.Equal(got, exp) { + t.Fatalf("expected %x got %x", exp, got) + } + if got, err := f.Retrieve(5); err != nil { + t.Fatal(err) + } else if exp := getChunk(20, 0xaa); !bytes.Equal(got, exp) { + t.Fatalf("expected %x got %x", exp, got) + } + if got, err := f.Retrieve(6); err != nil { + t.Fatal(err) + } else if exp := getChunk(20, 0x99); !bytes.Equal(got, exp) { + t.Fatalf("expected %x got %x", exp, got) + } + + // It should error at 0, 1,2,3 + for i := 0; i < 4; i++ { + if _, err := f.Retrieve(uint64(i)); err == nil { + t.Fatal("expected err") + } + } + } +} + +// TODO (?) +// - test that if we remove several head-files, aswell as data last data-file, +// the index is truncated accordingly +// Right now, the freezer would fail on these conditions: +// 1. have data files d0, d1, d2, d3 +// 2. remove d2,d3 +// +// However, all 'normal' failure modes arising due to failing to sync() or save a file should be +// handled already, and the case described above can only (?) happen if an external process/user +// deletes files from the filesystem. diff --git a/core/rawdb/schema.go b/core/rawdb/schema.go index 62b60e2f3..a44a2c99f 100644 --- a/core/rawdb/schema.go +++ b/core/rawdb/schema.go @@ -63,6 +63,33 @@ var ( preimageHitCounter = metrics.NewRegisteredCounter("db/preimage/hits", nil) ) +const ( + // freezerHeaderTable indicates the name of the freezer header table. + freezerHeaderTable = "headers" + + // freezerHashTable indicates the name of the freezer canonical hash table. + freezerHashTable = "hashes" + + // freezerBodiesTable indicates the name of the freezer block body table. + freezerBodiesTable = "bodies" + + // freezerReceiptTable indicates the name of the freezer receipts table. + freezerReceiptTable = "receipts" + + // freezerDifficultyTable indicates the name of the freezer total difficulty table. + freezerDifficultyTable = "diffs" +) + +// freezerNoSnappy configures whether compression is disabled for the ancient-tables. +// Hashes and difficulties don't compress well. +var freezerNoSnappy = map[string]bool{ + freezerHeaderTable: false, + freezerHashTable: true, + freezerBodiesTable: false, + freezerReceiptTable: false, + freezerDifficultyTable: true, +} + // LegacyTxLookupEntry is the legacy TxLookupEntry definition with some unnecessary // fields. type LegacyTxLookupEntry struct { diff --git a/core/rawdb/table.go b/core/rawdb/table.go index 0e50db7c9..6610b7f5a 100644 --- a/core/rawdb/table.go +++ b/core/rawdb/table.go @@ -50,6 +50,48 @@ func (t *table) Get(key []byte) ([]byte, error) { return t.db.Get(append([]byte(t.prefix), key...)) } +// HasAncient is a noop passthrough that just forwards the request to the underlying +// database. +func (t *table) HasAncient(kind string, number uint64) (bool, error) { + return t.db.HasAncient(kind, number) +} + +// Ancient is a noop passthrough that just forwards the request to the underlying +// database. +func (t *table) Ancient(kind string, number uint64) ([]byte, error) { + return t.db.Ancient(kind, number) +} + +// Ancients is a noop passthrough that just forwards the request to the underlying +// database. +func (t *table) Ancients() (uint64, error) { + return t.db.Ancients() +} + +// AncientSize is a noop passthrough that just forwards the request to the underlying +// database. +func (t *table) AncientSize(kind string) (uint64, error) { + return t.db.AncientSize(kind) +} + +// AppendAncient is a noop passthrough that just forwards the request to the underlying +// database. +func (t *table) AppendAncient(number uint64, hash, header, body, receipts, td []byte) error { + return t.db.AppendAncient(number, hash, header, body, receipts, td) +} + +// TruncateAncients is a noop passthrough that just forwards the request to the underlying +// database. +func (t *table) TruncateAncients(items uint64) error { + return t.db.TruncateAncients(items) +} + +// Sync is a noop passthrough that just forwards the request to the underlying +// database. +func (t *table) Sync() error { + return t.db.Sync() +} + // Put inserts the given value into the database at a prefixed version of the // provided key. func (t *table) Put(key []byte, value []byte) error { @@ -157,6 +199,6 @@ func (b *tableBatch) Reset() { } // Replay replays the batch contents. -func (b *tableBatch) Replay(w ethdb.Writer) error { +func (b *tableBatch) Replay(w ethdb.KeyValueWriter) error { return b.batch.Replay(w) } |