gaspersic/go-ethereum
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

core: initial version of state snapshots

Browse Source
This commit is contained in:
Péter Szilágyi
2019-08-06 13:40:28 +03:00
parent 2a5ed1a1d3
commit 542df8898e
30 changed files with 1635 additions and 85 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
core/state/snapshot/account.go Normal file
View File

@ -0,0 +1,54 @@
// 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 snapshot
import (
"bytes"
"math/big"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/rlp"
)
// Account is a slim version of a state.Account, where the root and code hash
// are replaced with a nil byte slice for empty accounts.
type Account struct {
Nonce uint64
Balance *big.Int
Root []byte
CodeHash []byte
}
// AccountRLP converts a state.Account content into a slim snapshot version RLP
// encoded.
func AccountRLP(nonce uint64, balance *big.Int, root common.Hash, codehash []byte) []byte {
slim := Account{
Nonce: nonce,
Balance: balance,
}
if root != emptyRoot {
slim.Root = root[:]
}
if !bytes.Equal(codehash, emptyCode[:]) {
slim.CodeHash = codehash
}
data, err := rlp.EncodeToBytes(slim)
if err != nil {
panic(err)
}
return data
}

337
core/state/snapshot/difflayer.go Normal file
View File

@ -0,0 +1,337 @@
// 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 snapshot
import (
"fmt"
"sort"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/rlp"
)
// diffLayer represents a collection of modifications made to a state snapshot
// after running a block on top. It contains one sorted list for the account trie
// and one-one list for each storage tries.
//
// The goal of a diff layer is to act as a journal, tracking recent modifications
// made to the state, that have not yet graduated into a semi-immutable state.
type diffLayer struct {
parent snapshot // Parent snapshot modified by this one, never nil
memory uint64 // Approximate guess as to how much memory we use
number uint64 // Block number to which this snapshot diff belongs to
root common.Hash // Root hash to which this snapshot diff belongs to
accountList []common.Hash // List of account for iteration, might not be sorted yet (lazy)
accountSorted bool // Flag whether the account list has alreayd been sorted or not
accountData map[common.Hash][]byte // Keyed accounts for direct retrival (nil means deleted)
storageList map[common.Hash][]common.Hash // List of storage slots for iterated retrievals, one per account
storageSorted map[common.Hash]bool // Flag whether the storage slot list has alreayd been sorted or not
storageData map[common.Hash]map[common.Hash][]byte // Keyed storage slots for direct retrival. one per account (nil means deleted)
lock sync.RWMutex
}
// newDiffLayer creates a new diff on top of an existing snapshot, whether that's a low
// level persistent database or a hierarchical diff already.
func newDiffLayer(parent snapshot, number uint64, root common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
// Create the new layer with some pre-allocated data segments
dl := &diffLayer{
parent: parent,
number: number,
root: root,
accountData: accounts,
storageData: storage,
}
// Fill the account hashes and sort them for the iterator
accountList := make([]common.Hash, 0, len(accounts))
for hash, data := range accounts {
accountList = append(accountList, hash)
dl.memory += uint64(len(data))
}
sort.Sort(hashes(accountList))
dl.accountList = accountList
dl.accountSorted = true
dl.memory += uint64(len(dl.accountList) * common.HashLength)
// Fill the storage hashes and sort them for the iterator
dl.storageList = make(map[common.Hash][]common.Hash, len(storage))
dl.storageSorted = make(map[common.Hash]bool, len(storage))
for accountHash, slots := range storage {
// If the slots are nil, sanity check that it's a deleted account
if slots == nil {
// Ensure that the account was just marked as deleted
if account, ok := accounts[accountHash]; account != nil || !ok {
panic(fmt.Sprintf("storage in %#x nil, but account conflicts (%#x, exists: %v)", accountHash, account, ok))
}
// Everything ok, store the deletion mark and continue
dl.storageList[accountHash] = nil
continue
}
// Storage slots are not nil so entire contract was not deleted, ensure the
// account was just updated.
if account, ok := accounts[accountHash]; account == nil || !ok {
log.Error(fmt.Sprintf("storage in %#x exists, but account nil (exists: %v)", accountHash, ok))
//panic(fmt.Sprintf("storage in %#x exists, but account nil (exists: %v)", accountHash, ok))
}
// Fill the storage hashes for this account and sort them for the iterator
storageList := make([]common.Hash, 0, len(slots))
for storageHash, data := range slots {
storageList = append(storageList, storageHash)
dl.memory += uint64(len(data))
}
sort.Sort(hashes(storageList))
dl.storageList[accountHash] = storageList
dl.storageSorted[accountHash] = true
dl.memory += uint64(len(storageList) * common.HashLength)
}
dl.memory += uint64(len(dl.storageList) * common.HashLength)
return dl
}
// Info returns the block number and root hash for which this snapshot was made.
func (dl *diffLayer) Info() (uint64, common.Hash) {
return dl.number, dl.root
}
// Account directly retrieves the account associated with a particular hash in
// the snapshot slim data format.
func (dl *diffLayer) Account(hash common.Hash) *Account {
data := dl.AccountRLP(hash)
if len(data) == 0 { // can be both nil and []byte{}
return nil
}
account := new(Account)
if err := rlp.DecodeBytes(data, account); err != nil {
panic(err)
}
return account
}
// AccountRLP directly retrieves the account RLP associated with a particular
// hash in the snapshot slim data format.
func (dl *diffLayer) AccountRLP(hash common.Hash) []byte {
dl.lock.RLock()
defer dl.lock.RUnlock()
// If the account is known locally, return it. Note, a nil account means it was
// deleted, and is a different notion than an unknown account!
if data, ok := dl.accountData[hash]; ok {
return data
}
// Account unknown to this diff, resolve from parent
return dl.parent.AccountRLP(hash)
}
// Storage directly retrieves the storage data associated with a particular hash,
// within a particular account. If the slot is unknown to this diff, it's parent
// is consulted.
func (dl *diffLayer) Storage(accountHash, storageHash common.Hash) []byte {
dl.lock.RLock()
defer dl.lock.RUnlock()
// If the account is known locally, try to resolve the slot locally. Note, a nil
// account means it was deleted, and is a different notion than an unknown account!
if storage, ok := dl.storageData[accountHash]; ok {
if storage == nil {
return nil
}
if data, ok := storage[storageHash]; ok {
return data
}
}
// Account - or slot within - unknown to this diff, resolve from parent
return dl.parent.Storage(accountHash, storageHash)
}
// Update creates a new layer on top of the existing snapshot diff tree with
// the specified data items.
func (dl *diffLayer) Update(blockRoot common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
return newDiffLayer(dl, dl.number+1, blockRoot, accounts, storage)
}
// Cap traverses downwards the diff tree until the number of allowed layers are
// crossed. All diffs beyond the permitted number are flattened downwards. If
// the layer limit is reached, memory cap is also enforced (but not before). The
// block numbers for the disk layer and first diff layer are returned for GC.
func (dl *diffLayer) Cap(layers int, memory uint64) (uint64, uint64) {
// Dive until we run out of layers or reach the persistent database
if layers > 2 {
// If we still have diff layers below, recurse
if parent, ok := dl.parent.(*diffLayer); ok {
return parent.Cap(layers-1, memory)
}
// Diff stack too shallow, return block numbers without modifications
return dl.parent.(*diskLayer).number, dl.number
}
// We're out of layers, flatten anything below, stopping if it's the disk or if
// the memory limit is not yet exceeded.
switch parent := dl.parent.(type) {
case *diskLayer:
return parent.number, dl.number
case *diffLayer:
dl.lock.Lock()
defer dl.lock.Unlock()
dl.parent = parent.flatten()
if dl.parent.(*diffLayer).memory < memory {
diskNumber, _ := parent.parent.Info()
return diskNumber, parent.number
}
default:
panic(fmt.Sprintf("unknown data layer: %T", parent))
}
// If the bottommost layer is larger than our memory cap, persist to disk
var (
parent = dl.parent.(*diffLayer)
base = parent.parent.(*diskLayer)
batch = base.db.NewBatch()
)
parent.lock.RLock()
defer parent.lock.RUnlock()
// Start by temporarilly deleting the current snapshot block marker. This
// ensures that in the case of a crash, the entire snapshot is invalidated.
rawdb.DeleteSnapshotBlock(batch)
// Push all the accounts into the database
for hash, data := range parent.accountData {
if len(data) > 0 {
// Account was updated, push to disk
rawdb.WriteAccountSnapshot(batch, hash, data)
base.cache.Set(string(hash[:]), data)
if batch.ValueSize() > ethdb.IdealBatchSize {
if err := batch.Write(); err != nil {
log.Crit("Failed to write account snapshot", "err", err)
}
batch.Reset()
}
} else {
// Account was deleted, remove all storage slots too
rawdb.DeleteAccountSnapshot(batch, hash)
base.cache.Set(string(hash[:]), nil)
it := rawdb.IterateStorageSnapshots(base.db, hash)
for it.Next() {
if key := it.Key(); len(key) == 65 { // TODO(karalabe): Yuck, we should move this into the iterator
batch.Delete(key)
base.cache.Delete(string(key[1:]))
}
}
it.Release()
}
}
// Push all the storage slots into the database
for accountHash, storage := range parent.storageData {
for storageHash, data := range storage {
if len(data) > 0 {
rawdb.WriteStorageSnapshot(batch, accountHash, storageHash, data)
base.cache.Set(string(append(accountHash[:], storageHash[:]...)), data)
} else {
rawdb.DeleteStorageSnapshot(batch, accountHash, storageHash)
base.cache.Set(string(append(accountHash[:], storageHash[:]...)), nil)
}
}
if batch.ValueSize() > ethdb.IdealBatchSize {
if err := batch.Write(); err != nil {
log.Crit("Failed to write storage snapshot", "err", err)
}
batch.Reset()
}
}
// Update the snapshot block marker and write any remainder data
base.number, base.root = parent.number, parent.root
rawdb.WriteSnapshotBlock(batch, base.number, base.root)
if err := batch.Write(); err != nil {
log.Crit("Failed to write leftover snapshot", "err", err)
}
dl.parent = base
return base.number, dl.number
}
// flatten pushes all data from this point downwards, flattening everything into
// a single diff at the bottom. Since usually the lowermost diff is the largest,
// the flattening bulds up from there in reverse.
func (dl *diffLayer) flatten() snapshot {
// If the parent is not diff, we're the first in line, return unmodified
parent, ok := dl.parent.(*diffLayer)
if !ok {
return dl
}
// Parent is a diff, flatten it first (note, apart from weird corned cases,
// flatten will realistically only ever merge 1 layer, so there's no need to
// be smarter about grouping flattens together).
parent = parent.flatten().(*diffLayer)
// Overwrite all the updated accounts blindly, merge the sorted list
for hash, data := range dl.accountData {
parent.accountData[hash] = data
}
parent.accountList = append(parent.accountList, dl.accountList...) // TODO(karalabe): dedup!!
parent.accountSorted = false
// Overwrite all the updates storage slots (individually)
for accountHash, storage := range dl.storageData {
// If storage didn't exist (or was deleted) in the parent; or if the storage
// was freshly deleted in the child, overwrite blindly
if parent.storageData[accountHash] == nil || storage == nil {
parent.storageList[accountHash] = dl.storageList[accountHash]
parent.storageData[accountHash] = storage
continue
}
// Storage exists in both parent and child, merge the slots
comboData := parent.storageData[accountHash]
for storageHash, data := range storage {
comboData[storageHash] = data
}
parent.storageData[accountHash] = comboData
parent.storageList[accountHash] = append(parent.storageList[accountHash], dl.storageList[accountHash]...) // TODO(karalabe): dedup!!
parent.storageSorted[accountHash] = false
}
// Return the combo parent
parent.number = dl.number
parent.root = dl.root
parent.memory += dl.memory
return parent
}
// Journal commits an entire diff hierarchy to disk into a single journal file.
// This is meant to be used during shutdown to persist the snapshot without
// flattening everything down (bad for reorgs).
func (dl *diffLayer) Journal() error {
dl.lock.RLock()
defer dl.lock.RUnlock()
writer, err := dl.journal()
if err != nil {
return err
}
writer.Close()
return nil
}

140
core/state/snapshot/difflayer_journal.go Normal file
View File

@ -0,0 +1,140 @@
// 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 snapshot
import (
"fmt"
"io"
"os"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/rlp"
)
// journalAccount is an account entry in a diffLayer's disk journal.
type journalAccount struct {
Hash common.Hash
Blob []byte
}
// journalStorage is an account's storage map in a diffLayer's disk journal.
type journalStorage struct {
Hash common.Hash
Keys []common.Hash
Vals [][]byte
}
// loadDiffLayer reads the next sections of a snapshot journal, reconstructing a new
// diff and verifying that it can be linked to the requested parent.
func loadDiffLayer(parent snapshot, r *rlp.Stream) (snapshot, error) {
// Read the next diff journal entry
var (
number uint64
root common.Hash
)
if err := r.Decode(&number); err != nil {
// The first read may fail with EOF, marking the end of the journal
if err == io.EOF {
return parent, nil
}
return nil, fmt.Errorf("load diff number: %v", err)
}
if err := r.Decode(&root); err != nil {
return nil, fmt.Errorf("load diff root: %v", err)
}
var accounts []journalAccount
if err := r.Decode(&accounts); err != nil {
return nil, fmt.Errorf("load diff accounts: %v", err)
}
accountData := make(map[common.Hash][]byte)
for _, entry := range accounts {
accountData[entry.Hash] = entry.Blob
}
var storage []journalStorage
if err := r.Decode(&storage); err != nil {
return nil, fmt.Errorf("load diff storage: %v", err)
}
storageData := make(map[common.Hash]map[common.Hash][]byte)
for _, entry := range storage {
slots := make(map[common.Hash][]byte)
for i, key := range entry.Keys {
slots[key] = entry.Vals[i]
}
storageData[entry.Hash] = slots
}
// Validate the block number to avoid state corruption
if parent, ok := parent.(*diffLayer); ok {
if number != parent.number+1 {
return nil, fmt.Errorf("snapshot chain broken: block #%d after #%d", number, parent.number)
}
}
return loadDiffLayer(newDiffLayer(parent, number, root, accountData, storageData), r)
}
// journal is the internal version of Journal that also returns the journal file
// so subsequent layers know where to write to.
func (dl *diffLayer) journal() (io.WriteCloser, error) {
// If we've reached the bottom, open the journal
var writer io.WriteCloser
if parent, ok := dl.parent.(*diskLayer); ok {
file, err := os.Create(parent.journal)
if err != nil {
return nil, err
}
writer = file
}
// If we haven't reached the bottom yet, journal the parent first
if writer == nil {
file, err := dl.parent.(*diffLayer).journal()
if err != nil {
return nil, err
}
writer = file
}
// Everything below was journalled, persist this layer too
if err := rlp.Encode(writer, dl.number); err != nil {
writer.Close()
return nil, err
}
if err := rlp.Encode(writer, dl.root); err != nil {
writer.Close()
return nil, err
}
accounts := make([]journalAccount, 0, len(dl.accountData))
for hash, blob := range dl.accountData {
accounts = append(accounts, journalAccount{Hash: hash, Blob: blob})
}
if err := rlp.Encode(writer, accounts); err != nil {
writer.Close()
return nil, err
}
storage := make([]journalStorage, 0, len(dl.storageData))
for hash, slots := range dl.storageData {
keys := make([]common.Hash, 0, len(slots))
vals := make([][]byte, 0, len(slots))
for key, val := range slots {
keys = append(keys, key)
vals = append(vals, val)
}
storage = append(storage, journalStorage{Hash: hash, Keys: keys, Vals: vals})
}
if err := rlp.Encode(writer, storage); err != nil {
writer.Close()
return nil, err
}
return writer, nil
}

115
core/state/snapshot/disklayer.go Normal file
View File

@ -0,0 +1,115 @@
// 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 snapshot
import (
"github.com/allegro/bigcache"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/rlp"
)
// diskLayer is a low level persistent snapshot built on top of a key-value store.
type diskLayer struct {
journal string // Path of the snapshot journal to use on shutdown
db ethdb.KeyValueStore // Key-value store containing the base snapshot
cache *bigcache.BigCache // Cache to avoid hitting the disk for direct access
number uint64 // Block number of the base snapshot
root common.Hash // Root hash of the base snapshot
}
// Info returns the block number and root hash for which this snapshot was made.
func (dl *diskLayer) Info() (uint64, common.Hash) {
return dl.number, dl.root
}
// Account directly retrieves the account associated with a particular hash in
// the snapshot slim data format.
func (dl *diskLayer) Account(hash common.Hash) *Account {
data := dl.AccountRLP(hash)
if len(data) == 0 { // can be both nil and []byte{}
return nil
}
account := new(Account)
if err := rlp.DecodeBytes(data, account); err != nil {
panic(err)
}
return account
}
// AccountRLP directly retrieves the account RLP associated with a particular
// hash in the snapshot slim data format.
func (dl *diskLayer) AccountRLP(hash common.Hash) []byte {
key := string(hash[:])
// Try to retrieve the account from the memory cache
if blob, err := dl.cache.Get(key); err == nil {
snapshotCleanHitMeter.Mark(1)
snapshotCleanReadMeter.Mark(int64(len(blob)))
return blob
}
// Cache doesn't contain account, pull from disk and cache for later
blob := rawdb.ReadAccountSnapshot(dl.db, hash)
dl.cache.Set(key, blob)
snapshotCleanMissMeter.Mark(1)
snapshotCleanWriteMeter.Mark(int64(len(blob)))
return blob
}
// Storage directly retrieves the storage data associated with a particular hash,
// within a particular account.
func (dl *diskLayer) Storage(accountHash, storageHash common.Hash) []byte {
key := string(append(accountHash[:], storageHash[:]...))
// Try to retrieve the storage slot from the memory cache
if blob, err := dl.cache.Get(key); err == nil {
snapshotCleanHitMeter.Mark(1)
snapshotCleanReadMeter.Mark(int64(len(blob)))
return blob
}
// Cache doesn't contain storage slot, pull from disk and cache for later
blob := rawdb.ReadStorageSnapshot(dl.db, accountHash, storageHash)
dl.cache.Set(key, blob)
snapshotCleanMissMeter.Mark(1)
snapshotCleanWriteMeter.Mark(int64(len(blob)))
return blob
}
// Update creates a new layer on top of the existing snapshot diff tree with
// the specified data items. Note, the maps are retained by the method to avoid
// copying everything.
func (dl *diskLayer) Update(blockHash common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
return newDiffLayer(dl, dl.number+1, blockHash, accounts, storage)
}
// Cap traverses downwards the diff tree until the number of allowed layers are
// crossed. All diffs beyond the permitted number are flattened downwards.
func (dl *diskLayer) Cap(layers int, memory uint64) (uint64, uint64) {
return dl.number, dl.number
}
// Journal commits an entire diff hierarchy to disk into a single journal file.
func (dl *diskLayer) Journal() error {
// There's no journalling a disk layer
return nil
}

212
core/state/snapshot/generate.go Normal file
View File

@ -0,0 +1,212 @@
// 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 snapshot
import (
"bytes"
"fmt"
"math/big"
"time"
"github.com/allegro/bigcache"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie"
)
var (
// emptyRoot is the known root hash of an empty trie.
emptyRoot = common.HexToHash("56e81f171bcc55a6ff8345e692c0f86e5b48e01b996cadc001622fb5e363b421")
// emptyCode is the known hash of the empty EVM bytecode.
emptyCode = crypto.Keccak256Hash(nil)
)
// wipeSnapshot iterates over the entire key-value database and deletes all the
// data associated with the snapshot (accounts, storage, metadata). After all is
// done, the snapshot range of the database is compacted to free up unused data
// blocks.
func wipeSnapshot(db ethdb.KeyValueStore) error {
// Batch deletions together to avoid holding an iterator for too long
var (
batch = db.NewBatch()
items int
)
// Iterate over the snapshot key-range and delete all of them
log.Info("Deleting previous snapshot leftovers")
start, logged := time.Now(), time.Now()
it := db.NewIteratorWithStart(rawdb.StateSnapshotPrefix)
for it.Next() {
// Skip any keys with the correct prefix but wrong lenth (trie nodes)
key := it.Key()
if !bytes.HasPrefix(key, rawdb.StateSnapshotPrefix) {
break
}
if len(key) != len(rawdb.StateSnapshotPrefix)+common.HashLength && len(key) != len(rawdb.StateSnapshotPrefix)+2*common.HashLength {
continue
}
// Delete the key and periodically recreate the batch and iterator
batch.Delete(key)
items++
if items%10000 == 0 {
// Batch too large (or iterator too long lived, flush and recreate)
it.Release()
if err := batch.Write(); err != nil {
return err
}
batch.Reset()
it = db.NewIteratorWithStart(key)
if time.Since(logged) > 8*time.Second {
log.Info("Deleting previous snapshot leftovers", "wiped", items, "elapsed", time.Since(start))
logged = time.Now()
}
}
}
it.Release()
rawdb.DeleteSnapshotBlock(batch)
if err := batch.Write(); err != nil {
return err
}
log.Info("Deleted previous snapshot leftovers", "wiped", items, "elapsed", time.Since(start))
// Compact the snapshot section of the database to get rid of unused space
log.Info("Compacting snapshot area in database")
start = time.Now()
end := common.CopyBytes(rawdb.StateSnapshotPrefix)
end[len(end)-1]++
if err := db.Compact(rawdb.StateSnapshotPrefix, end); err != nil {
return err
}
log.Info("Compacted snapshot area in database", "elapsed", time.Since(start))
return nil
}
// generateSnapshot regenerates a brand new snapshot based on an existing state database and head block.
func generateSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64, headRoot common.Hash) (snapshot, error) {
// Wipe any previously existing snapshot from the database
if err := wipeSnapshot(db); err != nil {
return nil, err
}
// Iterate the entire storage trie and re-generate the state snapshot
var (
accountCount int
storageCount int
storageNodes int
accountSize common.StorageSize
storageSize common.StorageSize
logged time.Time
)
batch := db.NewBatch()
triedb := trie.NewDatabase(db)
accTrie, err := trie.NewSecure(headRoot, triedb)
if err != nil {
return nil, err
}
accIt := trie.NewIterator(accTrie.NodeIterator(nil))
for accIt.Next() {
var (
curStorageCount int
curStorageNodes int
curAccountSize common.StorageSize
curStorageSize common.StorageSize
)
var acc struct {
Nonce uint64
Balance *big.Int
Root common.Hash
CodeHash []byte
}
if err := rlp.DecodeBytes(accIt.Value, &acc); err != nil {
return nil, err
}
data := AccountRLP(acc.Nonce, acc.Balance, acc.Root, acc.CodeHash)
curAccountSize += common.StorageSize(1 + common.HashLength + len(data))
rawdb.WriteAccountSnapshot(batch, common.BytesToHash(accIt.Key), data)
if batch.ValueSize() > ethdb.IdealBatchSize {
batch.Write()
batch.Reset()
}
if acc.Root != emptyRoot {
storeTrie, err := trie.NewSecure(acc.Root, triedb)
if err != nil {
return nil, err
}
storeIt := trie.NewIterator(storeTrie.NodeIterator(nil))
for storeIt.Next() {
curStorageSize += common.StorageSize(1 + 2*common.HashLength + len(storeIt.Value))
curStorageCount++
rawdb.WriteStorageSnapshot(batch, common.BytesToHash(accIt.Key), common.BytesToHash(storeIt.Key), storeIt.Value)
if batch.ValueSize() > ethdb.IdealBatchSize {
batch.Write()
batch.Reset()
}
}
curStorageNodes = storeIt.Nodes
}
accountCount++
storageCount += curStorageCount
accountSize += curAccountSize
storageSize += curStorageSize
storageNodes += curStorageNodes
if time.Since(logged) > 8*time.Second {
fmt.Printf("%#x: %9s + %9s (%6d slots, %6d nodes), total %9s (%d accs, %d nodes) + %9s (%d slots, %d nodes)\n", accIt.Key, curAccountSize.TerminalString(), curStorageSize.TerminalString(), curStorageCount, curStorageNodes, accountSize.TerminalString(), accountCount, accIt.Nodes, storageSize.TerminalString(), storageCount, storageNodes)
logged = time.Now()
}
}
fmt.Printf("Totals: %9s (%d accs, %d nodes) + %9s (%d slots, %d nodes)\n", accountSize.TerminalString(), accountCount, accIt.Nodes, storageSize.TerminalString(), storageCount, storageNodes)
// Update the snapshot block marker and write any remainder data
rawdb.WriteSnapshotBlock(batch, headNumber, headRoot)
batch.Write()
batch.Reset()
// Compact the snapshot section of the database to get rid of unused space
log.Info("Compacting snapshot in chain database")
if err := db.Compact([]byte{'s'}, []byte{'s' + 1}); err != nil {
return nil, err
}
// New snapshot generated, construct a brand new base layer
cache, _ := bigcache.NewBigCache(bigcache.Config{ // TODO(karalabe): dedup
Shards: 1024,
LifeWindow: time.Hour,
MaxEntriesInWindow: 512 * 1024,
MaxEntrySize: 512,
HardMaxCacheSize: 512,
})
return &diskLayer{
journal: journal,
db: db,
cache: cache,
number: headNumber,
root: headRoot,
}, nil
}

111
core/state/snapshot/generate_test.go Normal file
View File

@ -0,0 +1,111 @@
// 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 snapshot
import (
"math/rand"
"testing"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb/memorydb"
)
// randomHash generates a random blob of data and returns it as a hash.
func randomHash() common.Hash {
var hash common.Hash
if n, err := rand.Read(hash[:]); n != common.HashLength || err != nil {
panic(err)
}
return hash
}
// Tests that given a database with random data content, all parts of a snapshot
// can be crrectly wiped without touching anything else.
func TestWipe(t *testing.T) {
// Create a database with some random snapshot data
db := memorydb.New()
for i := 0; i < 128; i++ {
account := randomHash()
rawdb.WriteAccountSnapshot(db, account, randomHash().Bytes())
for j := 0; j < 1024; j++ {
rawdb.WriteStorageSnapshot(db, account, randomHash(), randomHash().Bytes())
}
}
rawdb.WriteSnapshotBlock(db, 123, randomHash())
// Add some random non-snapshot data too to make wiping harder
for i := 0; i < 65536; i++ {
// Generate a key that's the wrong length for a state snapshot item
var keysize int
for keysize == 0 || keysize == 32 || keysize == 64 {
keysize = 8 + rand.Intn(64) // +8 to ensure we will "never" randomize duplicates
}
// Randomize the suffix, dedup and inject it under the snapshot namespace
keysuffix := make([]byte, keysize)
rand.Read(keysuffix)
db.Put(append(rawdb.StateSnapshotPrefix, keysuffix...), randomHash().Bytes())
}
// Sanity check that all the keys are present
var items int
it := db.NewIteratorWithPrefix(rawdb.StateSnapshotPrefix)
defer it.Release()
for it.Next() {
key := it.Key()
if len(key) == len(rawdb.StateSnapshotPrefix)+32 || len(key) == len(rawdb.StateSnapshotPrefix)+64 {
items++
}
}
if items != 128+128*1024 {
t.Fatalf("snapshot size mismatch: have %d, want %d", items, 128+128*1024)
}
if number, hash := rawdb.ReadSnapshotBlock(db); number != 123 || hash == (common.Hash{}) {
t.Errorf("snapshot block marker mismatch: have #%d [%#x], want #%d [<not-nil>]", number, hash, 123)
}
// Wipe all snapshot entries from the database
if err := wipeSnapshot(db); err != nil {
t.Fatalf("failed to wipe snapshot: %v", err)
}
// Iterate over the database end ensure no snapshot information remains
it = db.NewIteratorWithPrefix(rawdb.StateSnapshotPrefix)
defer it.Release()
for it.Next() {
key := it.Key()
if len(key) == len(rawdb.StateSnapshotPrefix)+32 || len(key) == len(rawdb.StateSnapshotPrefix)+64 {
t.Errorf("snapshot entry remained after wipe: %x", key)
}
}
if number, hash := rawdb.ReadSnapshotBlock(db); number != 0 || hash != (common.Hash{}) {
t.Errorf("snapshot block marker remained after wipe: #%d [%#x]", number, hash)
}
// Iterate over the database and ensure miscellaneous items are present
items = 0
it = db.NewIterator()
defer it.Release()
for it.Next() {
items++
}
if items != 65536 {
t.Fatalf("misc item count mismatch: have %d, want %d", items, 65536)
}
}

244
core/state/snapshot/snapshot.go Normal file
View File

@ -0,0 +1,244 @@
// 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 snapshot implements a journalled, dynamic state dump.
package snapshot
import (
"errors"
"fmt"
"os"
"sync"
"time"
"github.com/allegro/bigcache"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/metrics"
"github.com/ethereum/go-ethereum/rlp"
)
var (
snapshotCleanHitMeter = metrics.NewRegisteredMeter("state/snapshot/clean/hit", nil)
snapshotCleanMissMeter = metrics.NewRegisteredMeter("state/snapshot/clean/miss", nil)
snapshotCleanReadMeter = metrics.NewRegisteredMeter("state/snapshot/clean/read", nil)
snapshotCleanWriteMeter = metrics.NewRegisteredMeter("state/snapshot/clean/write", nil)
)
// Snapshot represents the functionality supported by a snapshot storage layer.
type Snapshot interface {
// Info returns the block number and root hash for which this snapshot was made.
Info() (uint64, common.Hash)
// Account directly retrieves the account associated with a particular hash in
// the snapshot slim data format.
Account(hash common.Hash) *Account
// AccountRLP directly retrieves the account RLP associated with a particular
// hash in the snapshot slim data format.
AccountRLP(hash common.Hash) []byte
// Storage directly retrieves the storage data associated with a particular hash,
// within a particular account.
Storage(accountHash, storageHash common.Hash) []byte
}
// snapshot is the internal version of the snapshot data layer that supports some
// additional methods compared to the public API.
type snapshot interface {
Snapshot
// Update creates a new layer on top of the existing snapshot diff tree with
// the specified data items. Note, the maps are retained by the method to avoid
// copying everything.
Update(blockRoot common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer
// Cap traverses downwards the diff tree until the number of allowed layers are
// crossed. All diffs beyond the permitted number are flattened downwards. The
// block numbers for the disk layer and first diff layer are returned for GC.
Cap(layers int, memory uint64) (uint64, uint64)
// Journal commits an entire diff hierarchy to disk into a single journal file.
// This is meant to be used during shutdown to persist the snapshot without
// flattening everything down (bad for reorgs).
Journal() error
}
// SnapshotTree is an Ethereum state snapshot tree. It consists of one persistent
// base layer backed by a key-value store, on top of which arbitrarilly many in-
// memory diff layers are topped. The memory diffs can form a tree with branching,
// but the disk layer is singleton and common to all. If a reorg goes deeper than
// the disk layer, everything needs to be deleted.
//
// The goal of a state snapshot is twofold: to allow direct access to account and
// storage data to avoid expensive multi-level trie lookups; and to allow sorted,
// cheap iteration of the account/storage tries for sync aid.
type SnapshotTree struct {
layers map[common.Hash]snapshot // Collection of all known layers // TODO(karalabe): split Clique overlaps
lock sync.RWMutex
}
// New attempts to load an already existing snapshot from a persistent key-value
// store (with a number of memory layers from a journal), ensuring that the head
// of the snapshot matches the expected one.
//
// If the snapshot is missing or inconsistent, the entirety is deleted and will
// be reconstructed from scratch based on the tries in the key-value store.
func New(db ethdb.KeyValueStore, journal string, headNumber uint64, headRoot common.Hash) (*SnapshotTree, error) {
// Attempt to load a previously persisted snapshot
head, err := loadSnapshot(db, journal, headNumber, headRoot)
if err != nil {
log.Warn("Failed to load snapshot, regenerating", "err", err)
if head, err = generateSnapshot(db, journal, headNumber, headRoot); err != nil {
return nil, err
}
}
// Existing snapshot loaded or one regenerated, seed all the layers
snap := &SnapshotTree{
layers: make(map[common.Hash]snapshot),
}
for head != nil {
_, root := head.Info()
snap.layers[root] = head
switch self := head.(type) {
case *diffLayer:
head = self.parent
case *diskLayer:
head = nil
default:
panic(fmt.Sprintf("unknown data layer: %T", self))
}
}
return snap, nil
}
// Snapshot retrieves a snapshot belonging to the given block root, or nil if no
// snapshot is maintained for that block.
func (st *SnapshotTree) Snapshot(blockRoot common.Hash) Snapshot {
st.lock.RLock()
defer st.lock.RUnlock()
return st.layers[blockRoot]
}
// Update adds a new snapshot into the tree, if that can be linked to an existing
// old parent. It is disallowed to insert a disk layer (the origin of all).
func (st *SnapshotTree) Update(blockRoot common.Hash, parentRoot common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) error {
// Generate a new snapshot on top of the parent
parent := st.Snapshot(parentRoot).(snapshot)
if parent == nil {
return fmt.Errorf("parent [%#x] snapshot missing", parentRoot)
}
snap := parent.Update(blockRoot, accounts, storage)
// Save the new snapshot for later
st.lock.Lock()
defer st.lock.Unlock()
st.layers[snap.root] = snap
return nil
}
// Cap traverses downwards the snapshot tree from a head block hash until the
// number of allowed layers are crossed. All layers beyond the permitted number
// are flattened downwards.
func (st *SnapshotTree) Cap(blockRoot common.Hash, layers int, memory uint64) error {
// Retrieve the head snapshot to cap from
snap := st.Snapshot(blockRoot).(snapshot)
if snap == nil {
return fmt.Errorf("snapshot [%#x] missing", blockRoot)
}
// Run the internal capping and discard all stale layers
st.lock.Lock()
defer st.lock.Unlock()
diskNumber, diffNumber := snap.Cap(layers, memory)
for root, snap := range st.layers {
if number, _ := snap.Info(); number != diskNumber && number < diffNumber {
delete(st.layers, root)
}
}
return nil
}
// Journal commits an entire diff hierarchy to disk into a single journal file.
// This is meant to be used during shutdown to persist the snapshot without
// flattening everything down (bad for reorgs).
func (st *SnapshotTree) Journal(blockRoot common.Hash) error {
// Retrieve the head snapshot to journal from
snap := st.Snapshot(blockRoot).(snapshot)
if snap == nil {
return fmt.Errorf("snapshot [%#x] missing", blockRoot)
}
// Run the journaling
st.lock.Lock()
defer st.lock.Unlock()
return snap.Journal()
}
// loadSnapshot loads a pre-existing state snapshot backed by a key-value store.
func loadSnapshot(db ethdb.KeyValueStore, journal string, headNumber uint64, headRoot common.Hash) (snapshot, error) {
// Retrieve the block number and hash of the snapshot, failing if no snapshot
// is present in the database (or crashed mid-update).
number, root := rawdb.ReadSnapshotBlock(db)
if root == (common.Hash{}) {
return nil, errors.New("missing or corrupted snapshot")
}
cache, _ := bigcache.NewBigCache(bigcache.Config{ // TODO(karalabe): dedup
Shards: 1024,
LifeWindow: time.Hour,
MaxEntriesInWindow: 512 * 1024,
MaxEntrySize: 512,
HardMaxCacheSize: 512,
})
base := &diskLayer{
journal: journal,
db: db,
cache: cache,
number: number,
root: root,
}
// Load all the snapshot diffs from the journal, failing if their chain is broken
// or does not lead from the disk snapshot to the specified head.
if _, err := os.Stat(journal); os.IsNotExist(err) {
// Journal doesn't exist, don't worry if it's not supposed to
if number != headNumber || root != headRoot {
return nil, fmt.Errorf("snapshot journal missing, head does't match snapshot: #%d [%#x] vs. #%d [%#x]",
headNumber, headRoot, number, root)
}
return base, nil
}
file, err := os.Open(journal)
if err != nil {
return nil, err
}
snapshot, err := loadDiffLayer(base, rlp.NewStream(file, 0))
if err != nil {
return nil, err
}
// Entire snapshot journal loaded, sanity check the head and return
// Journal doesn't exist, don't worry if it's not supposed to
number, root = snapshot.Info()
if number != headNumber || root != headRoot {
return nil, fmt.Errorf("head does't match snapshot: #%d [%#x] vs. #%d [%#x]",
headNumber, headRoot, number, root)
}
return snapshot, nil
}

17
core/state/snapshot/snapshot_test.go Normal file
View File

@ -0,0 +1,17 @@
// 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 snapshot

62
core/state/snapshot/sort.go Normal file
View File

@ -0,0 +1,62 @@
// 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 snapshot
import (
"bytes"
"github.com/ethereum/go-ethereum/common"
)
// hashes is a helper to implement sort.Interface.
type hashes []common.Hash
// Len is the number of elements in the collection.
func (hs hashes) Len() int { return len(hs) }
// Less reports whether the element with index i should sort before the element
// with index j.
func (hs hashes) Less(i, j int) bool { return bytes.Compare(hs[i][:], hs[j][:]) < 0 }
// Swap swaps the elements with indexes i and j.
func (hs hashes) Swap(i, j int) { hs[i], hs[j] = hs[j], hs[i] }
// merge combines two sorted lists of hashes into a combo sorted one.
func merge(a, b []common.Hash) []common.Hash {
result := make([]common.Hash, len(a)+len(b))
i := 0
for len(a) > 0 && len(b) > 0 {
if bytes.Compare(a[0][:], b[0][:]) < 0 {
result[i] = a[0]
a = a[1:]
} else {
result[i] = b[0]
b = b[1:]
}
i++
}
for j := 0; j < len(a); j++ {
result[i] = a[j]
i++
}
for j := 0; j < len(b); j++ {
result[i] = b[j]
i++
}
return result
}