// 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 . package snapshot import ( "bytes" "errors" "fmt" "time" "github.com/VictoriaMetrics/fastcache" "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/common/hexutil" "github.com/ethereum/go-ethereum/core/rawdb" "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/ethdb" "github.com/ethereum/go-ethereum/log" "github.com/ethereum/go-ethereum/rlp" "github.com/ethereum/go-ethereum/trie" "github.com/ethereum/go-ethereum/trie/trienode" ) var ( // accountCheckRange is the upper limit of the number of accounts involved in // each range check. This is a value estimated based on experience. If this // range is too large, the failure rate of range proof will increase. Otherwise, // if the range is too small, the efficiency of the state recovery will decrease. accountCheckRange = 128 // storageCheckRange is the upper limit of the number of storage slots involved // in each range check. This is a value estimated based on experience. If this // range is too large, the failure rate of range proof will increase. Otherwise, // if the range is too small, the efficiency of the state recovery will decrease. storageCheckRange = 1024 // errMissingTrie is returned if the target trie is missing while the generation // is running. In this case the generation is aborted and wait the new signal. errMissingTrie = errors.New("missing trie") ) // generateSnapshot regenerates a brand new snapshot based on an existing state // database and head block asynchronously. The snapshot is returned immediately // and generation is continued in the background until done. func generateSnapshot(diskdb ethdb.KeyValueStore, triedb *trie.Database, cache int, root common.Hash) *diskLayer { // Create a new disk layer with an initialized state marker at zero var ( stats = &generatorStats{start: time.Now()} batch = diskdb.NewBatch() genMarker = []byte{} // Initialized but empty! ) rawdb.WriteSnapshotRoot(batch, root) journalProgress(batch, genMarker, stats) if err := batch.Write(); err != nil { log.Crit("Failed to write initialized state marker", "err", err) } base := &diskLayer{ diskdb: diskdb, triedb: triedb, root: root, cache: fastcache.New(cache * 1024 * 1024), genMarker: genMarker, genPending: make(chan struct{}), genAbort: make(chan chan *generatorStats), } go base.generate(stats) log.Debug("Start snapshot generation", "root", root) return base } // journalProgress persists the generator stats into the database to resume later. func journalProgress(db ethdb.KeyValueWriter, marker []byte, stats *generatorStats) { // Write out the generator marker. Note it's a standalone disk layer generator // which is not mixed with journal. It's ok if the generator is persisted while // journal is not. entry := journalGenerator{ Done: marker == nil, Marker: marker, } if stats != nil { entry.Accounts = stats.accounts entry.Slots = stats.slots entry.Storage = uint64(stats.storage) } blob, err := rlp.EncodeToBytes(entry) if err != nil { panic(err) // Cannot happen, here to catch dev errors } var logstr string switch { case marker == nil: logstr = "done" case bytes.Equal(marker, []byte{}): logstr = "empty" case len(marker) == common.HashLength: logstr = fmt.Sprintf("%#x", marker) default: logstr = fmt.Sprintf("%#x:%#x", marker[:common.HashLength], marker[common.HashLength:]) } log.Debug("Journaled generator progress", "progress", logstr) rawdb.WriteSnapshotGenerator(db, blob) } // proofResult contains the output of range proving which can be used // for further processing regardless if it is successful or not. type proofResult struct { keys [][]byte // The key set of all elements being iterated, even proving is failed vals [][]byte // The val set of all elements being iterated, even proving is failed diskMore bool // Set when the database has extra snapshot states since last iteration trieMore bool // Set when the trie has extra snapshot states(only meaningful for successful proving) proofErr error // Indicator whether the given state range is valid or not tr *trie.Trie // The trie, in case the trie was resolved by the prover (may be nil) } // valid returns the indicator that range proof is successful or not. func (result *proofResult) valid() bool { return result.proofErr == nil } // last returns the last verified element key regardless of whether the range proof is // successful or not. Nil is returned if nothing involved in the proving. func (result *proofResult) last() []byte { var last []byte if len(result.keys) > 0 { last = result.keys[len(result.keys)-1] } return last } // forEach iterates all the visited elements and applies the given callback on them. // The iteration is aborted if the callback returns non-nil error. func (result *proofResult) forEach(callback func(key []byte, val []byte) error) error { for i := 0; i < len(result.keys); i++ { key, val := result.keys[i], result.vals[i] if err := callback(key, val); err != nil { return err } } return nil } // proveRange proves the snapshot segment with particular prefix is "valid". // The iteration start point will be assigned if the iterator is restored from // the last interruption. Max will be assigned in order to limit the maximum // amount of data involved in each iteration. // // The proof result will be returned if the range proving is finished, otherwise // the error will be returned to abort the entire procedure. func (dl *diskLayer) proveRange(ctx *generatorContext, trieId *trie.ID, prefix []byte, kind string, origin []byte, max int, valueConvertFn func([]byte) ([]byte, error)) (*proofResult, error) { var ( keys [][]byte vals [][]byte proof = rawdb.NewMemoryDatabase() diskMore = false iter = ctx.iterator(kind) start = time.Now() min = append(prefix, origin...) ) for iter.Next() { // Ensure the iterated item is always equal or larger than the given origin. key := iter.Key() if bytes.Compare(key, min) < 0 { return nil, errors.New("invalid iteration position") } // Ensure the iterated item still fall in the specified prefix. If // not which means the items in the specified area are all visited. // Move the iterator a step back since we iterate one extra element // out. if !bytes.Equal(key[:len(prefix)], prefix) { iter.Hold() break } // Break if we've reached the max size, and signal that we're not // done yet. Move the iterator a step back since we iterate one // extra element out. if len(keys) == max { iter.Hold() diskMore = true break } keys = append(keys, common.CopyBytes(key[len(prefix):])) if valueConvertFn == nil { vals = append(vals, common.CopyBytes(iter.Value())) } else { val, err := valueConvertFn(iter.Value()) if err != nil { // Special case, the state data is corrupted (invalid slim-format account), // don't abort the entire procedure directly. Instead, let the fallback // generation to heal the invalid data. // // Here append the original value to ensure that the number of key and // value are aligned. vals = append(vals, common.CopyBytes(iter.Value())) log.Error("Failed to convert account state data", "err", err) } else { vals = append(vals, val) } } } // Update metrics for database iteration and merkle proving if kind == snapStorage { snapStorageSnapReadCounter.Inc(time.Since(start).Nanoseconds()) } else { snapAccountSnapReadCounter.Inc(time.Since(start).Nanoseconds()) } defer func(start time.Time) { if kind == snapStorage { snapStorageProveCounter.Inc(time.Since(start).Nanoseconds()) } else { snapAccountProveCounter.Inc(time.Since(start).Nanoseconds()) } }(time.Now()) // The snap state is exhausted, pass the entire key/val set for verification root := trieId.Root if origin == nil && !diskMore { stackTr := trie.NewStackTrie(nil) for i, key := range keys { stackTr.Update(key, vals[i]) } if gotRoot := stackTr.Hash(); gotRoot != root { return &proofResult{ keys: keys, vals: vals, proofErr: fmt.Errorf("wrong root: have %#x want %#x", gotRoot, root), }, nil } return &proofResult{keys: keys, vals: vals}, nil } // Snap state is chunked, generate edge proofs for verification. tr, err := trie.New(trieId, dl.triedb) if err != nil { ctx.stats.Log("Trie missing, state snapshotting paused", dl.root, dl.genMarker) return nil, errMissingTrie } // Firstly find out the key of last iterated element. var last []byte if len(keys) > 0 { last = keys[len(keys)-1] } // Generate the Merkle proofs for the first and last element if origin == nil { origin = common.Hash{}.Bytes() } if err := tr.Prove(origin, proof); err != nil { log.Debug("Failed to prove range", "kind", kind, "origin", origin, "err", err) return &proofResult{ keys: keys, vals: vals, diskMore: diskMore, proofErr: err, tr: tr, }, nil } if last != nil { if err := tr.Prove(last, proof); err != nil { log.Debug("Failed to prove range", "kind", kind, "last", last, "err", err) return &proofResult{ keys: keys, vals: vals, diskMore: diskMore, proofErr: err, tr: tr, }, nil } } // Verify the snapshot segment with range prover, ensure that all flat states // in this range correspond to merkle trie. cont, err := trie.VerifyRangeProof(root, origin, last, keys, vals, proof) return &proofResult{ keys: keys, vals: vals, diskMore: diskMore, trieMore: cont, proofErr: err, tr: tr}, nil } // onStateCallback is a function that is called by generateRange, when processing a range of // accounts or storage slots. For each element, the callback is invoked. // // - If 'delete' is true, then this element (and potential slots) needs to be deleted from the snapshot. // - If 'write' is true, then this element needs to be updated with the 'val'. // - If 'write' is false, then this element is already correct, and needs no update. // The 'val' is the canonical encoding of the value (not the slim format for accounts) // // However, for accounts, the storage trie of the account needs to be checked. Also, // dangling storages(storage exists but the corresponding account is missing) need to // be cleaned up. type onStateCallback func(key []byte, val []byte, write bool, delete bool) error // generateRange generates the state segment with particular prefix. Generation can // either verify the correctness of existing state through range-proof and skip // generation, or iterate trie to regenerate state on demand. func (dl *diskLayer) generateRange(ctx *generatorContext, trieId *trie.ID, prefix []byte, kind string, origin []byte, max int, onState onStateCallback, valueConvertFn func([]byte) ([]byte, error)) (bool, []byte, error) { // Use range prover to check the validity of the flat state in the range result, err := dl.proveRange(ctx, trieId, prefix, kind, origin, max, valueConvertFn) if err != nil { return false, nil, err } last := result.last() // Construct contextual logger logCtx := []interface{}{"kind", kind, "prefix", hexutil.Encode(prefix)} if len(origin) > 0 { logCtx = append(logCtx, "origin", hexutil.Encode(origin)) } logger := log.New(logCtx...) // The range prover says the range is correct, skip trie iteration if result.valid() { snapSuccessfulRangeProofMeter.Mark(1) logger.Trace("Proved state range", "last", hexutil.Encode(last)) // The verification is passed, process each state with the given // callback function. If this state represents a contract, the // corresponding storage check will be performed in the callback if err := result.forEach(func(key []byte, val []byte) error { return onState(key, val, false, false) }); err != nil { return false, nil, err } // Only abort the iteration when both database and trie are exhausted return !result.diskMore && !result.trieMore, last, nil } logger.Trace("Detected outdated state range", "last", hexutil.Encode(last), "err", result.proofErr) snapFailedRangeProofMeter.Mark(1) // Special case, the entire trie is missing. In the original trie scheme, // all the duplicated subtries will be filtered out (only one copy of data // will be stored). While in the snapshot model, all the storage tries // belong to different contracts will be kept even they are duplicated. // Track it to a certain extent remove the noise data used for statistics. if origin == nil && last == nil { meter := snapMissallAccountMeter if kind == snapStorage { meter = snapMissallStorageMeter } meter.Mark(1) } // We use the snap data to build up a cache which can be used by the // main account trie as a primary lookup when resolving hashes var resolver trie.NodeResolver if len(result.keys) > 0 { mdb := rawdb.NewMemoryDatabase() tdb := trie.NewDatabase(mdb, trie.HashDefaults) defer tdb.Close() snapTrie := trie.NewEmpty(tdb) for i, key := range result.keys { snapTrie.Update(key, result.vals[i]) } root, nodes, err := snapTrie.Commit(false) if err != nil { return false, nil, err } if nodes != nil { tdb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil) tdb.Commit(root, false) } resolver = func(owner common.Hash, path []byte, hash common.Hash) []byte { return rawdb.ReadTrieNode(mdb, owner, path, hash, tdb.Scheme()) } } // Construct the trie for state iteration, reuse the trie // if it's already opened with some nodes resolved. tr := result.tr if tr == nil { tr, err = trie.New(trieId, dl.triedb) if err != nil { ctx.stats.Log("Trie missing, state snapshotting paused", dl.root, dl.genMarker) return false, nil, errMissingTrie } } var ( trieMore bool kvkeys, kvvals = result.keys, result.vals // counters count = 0 // number of states delivered by iterator created = 0 // states created from the trie updated = 0 // states updated from the trie deleted = 0 // states not in trie, but were in snapshot untouched = 0 // states already correct // timers start = time.Now() internal time.Duration ) nodeIt, err := tr.NodeIterator(origin) if err != nil { return false, nil, err } nodeIt.AddResolver(resolver) iter := trie.NewIterator(nodeIt) for iter.Next() { if last != nil && bytes.Compare(iter.Key, last) > 0 { trieMore = true break } count++ write := true created++ for len(kvkeys) > 0 { if cmp := bytes.Compare(kvkeys[0], iter.Key); cmp < 0 { // delete the key istart := time.Now() if err := onState(kvkeys[0], nil, false, true); err != nil { return false, nil, err } kvkeys = kvkeys[1:] kvvals = kvvals[1:] deleted++ internal += time.Since(istart) continue } else if cmp == 0 { // the snapshot key can be overwritten created-- if write = !bytes.Equal(kvvals[0], iter.Value); write { updated++ } else { untouched++ } kvkeys = kvkeys[1:] kvvals = kvvals[1:] } break } istart := time.Now() if err := onState(iter.Key, iter.Value, write, false); err != nil { return false, nil, err } internal += time.Since(istart) } if iter.Err != nil { return false, nil, iter.Err } // Delete all stale snapshot states remaining istart := time.Now() for _, key := range kvkeys { if err := onState(key, nil, false, true); err != nil { return false, nil, err } deleted += 1 } internal += time.Since(istart) // Update metrics for counting trie iteration if kind == snapStorage { snapStorageTrieReadCounter.Inc((time.Since(start) - internal).Nanoseconds()) } else { snapAccountTrieReadCounter.Inc((time.Since(start) - internal).Nanoseconds()) } logger.Debug("Regenerated state range", "root", trieId.Root, "last", hexutil.Encode(last), "count", count, "created", created, "updated", updated, "untouched", untouched, "deleted", deleted) // If there are either more trie items, or there are more snap items // (in the next segment), then we need to keep working return !trieMore && !result.diskMore, last, nil } // checkAndFlush checks if an interruption signal is received or the // batch size has exceeded the allowance. func (dl *diskLayer) checkAndFlush(ctx *generatorContext, current []byte) error { var abort chan *generatorStats select { case abort = <-dl.genAbort: default: } if ctx.batch.ValueSize() > ethdb.IdealBatchSize || abort != nil { if bytes.Compare(current, dl.genMarker) < 0 { log.Error("Snapshot generator went backwards", "current", fmt.Sprintf("%x", current), "genMarker", fmt.Sprintf("%x", dl.genMarker)) } // Flush out the batch anyway no matter it's empty or not. // It's possible that all the states are recovered and the // generation indeed makes progress. journalProgress(ctx.batch, current, ctx.stats) if err := ctx.batch.Write(); err != nil { return err } ctx.batch.Reset() dl.lock.Lock() dl.genMarker = current dl.lock.Unlock() if abort != nil { ctx.stats.Log("Aborting state snapshot generation", dl.root, current) return newAbortErr(abort) // bubble up an error for interruption } // Don't hold the iterators too long, release them to let compactor works ctx.reopenIterator(snapAccount) ctx.reopenIterator(snapStorage) } if time.Since(ctx.logged) > 8*time.Second { ctx.stats.Log("Generating state snapshot", dl.root, current) ctx.logged = time.Now() } return nil } // generateStorages generates the missing storage slots of the specific contract. // It's supposed to restart the generation from the given origin position. func generateStorages(ctx *generatorContext, dl *diskLayer, stateRoot common.Hash, account common.Hash, storageRoot common.Hash, storeMarker []byte) error { onStorage := func(key []byte, val []byte, write bool, delete bool) error { defer func(start time.Time) { snapStorageWriteCounter.Inc(time.Since(start).Nanoseconds()) }(time.Now()) if delete { rawdb.DeleteStorageSnapshot(ctx.batch, account, common.BytesToHash(key)) snapWipedStorageMeter.Mark(1) return nil } if write { rawdb.WriteStorageSnapshot(ctx.batch, account, common.BytesToHash(key), val) snapGeneratedStorageMeter.Mark(1) } else { snapRecoveredStorageMeter.Mark(1) } ctx.stats.storage += common.StorageSize(1 + 2*common.HashLength + len(val)) ctx.stats.slots++ // If we've exceeded our batch allowance or termination was requested, flush to disk if err := dl.checkAndFlush(ctx, append(account[:], key...)); err != nil { return err } return nil } // Loop for re-generating the missing storage slots. var origin = common.CopyBytes(storeMarker) for { id := trie.StorageTrieID(stateRoot, account, storageRoot) exhausted, last, err := dl.generateRange(ctx, id, append(rawdb.SnapshotStoragePrefix, account.Bytes()...), snapStorage, origin, storageCheckRange, onStorage, nil) if err != nil { return err // The procedure it aborted, either by external signal or internal error. } // Abort the procedure if the entire contract storage is generated if exhausted { break } if origin = increaseKey(last); origin == nil { break // special case, the last is 0xffffffff...fff } } return nil } // generateAccounts generates the missing snapshot accounts as well as their // storage slots in the main trie. It's supposed to restart the generation // from the given origin position. func generateAccounts(ctx *generatorContext, dl *diskLayer, accMarker []byte) error { onAccount := func(key []byte, val []byte, write bool, delete bool) error { // Make sure to clear all dangling storages before this account account := common.BytesToHash(key) ctx.removeStorageBefore(account) start := time.Now() if delete { rawdb.DeleteAccountSnapshot(ctx.batch, account) snapWipedAccountMeter.Mark(1) snapAccountWriteCounter.Inc(time.Since(start).Nanoseconds()) ctx.removeStorageAt(account) return nil } // Retrieve the current account and flatten it into the internal format var acc types.StateAccount if err := rlp.DecodeBytes(val, &acc); err != nil { log.Crit("Invalid account encountered during snapshot creation", "err", err) } // If the account is not yet in-progress, write it out if accMarker == nil || !bytes.Equal(account[:], accMarker) { dataLen := len(val) // Approximate size, saves us a round of RLP-encoding if !write { if bytes.Equal(acc.CodeHash, types.EmptyCodeHash[:]) { dataLen -= 32 } if acc.Root == types.EmptyRootHash { dataLen -= 32 } snapRecoveredAccountMeter.Mark(1) } else { data := types.SlimAccountRLP(acc) dataLen = len(data) rawdb.WriteAccountSnapshot(ctx.batch, account, data) snapGeneratedAccountMeter.Mark(1) } ctx.stats.storage += common.StorageSize(1 + common.HashLength + dataLen) ctx.stats.accounts++ } // If the snap generation goes here after interrupted, genMarker may go backward // when last genMarker is consisted of accountHash and storageHash marker := account[:] if accMarker != nil && bytes.Equal(marker, accMarker) && len(dl.genMarker) > common.HashLength { marker = dl.genMarker[:] } // If we've exceeded our batch allowance or termination was requested, flush to disk if err := dl.checkAndFlush(ctx, marker); err != nil { return err } snapAccountWriteCounter.Inc(time.Since(start).Nanoseconds()) // let's count flush time as well // If the iterated account is the contract, create a further loop to // verify or regenerate the contract storage. if acc.Root == types.EmptyRootHash { ctx.removeStorageAt(account) } else { var storeMarker []byte if accMarker != nil && bytes.Equal(account[:], accMarker) && len(dl.genMarker) > common.HashLength { storeMarker = dl.genMarker[common.HashLength:] } if err := generateStorages(ctx, dl, dl.root, account, acc.Root, storeMarker); err != nil { return err } } // Some account processed, unmark the marker accMarker = nil return nil } // Always reset the initial account range as 1 whenever recover from the // interruption. TODO(rjl493456442) can we remove it? var accountRange = accountCheckRange if len(accMarker) > 0 { accountRange = 1 } origin := common.CopyBytes(accMarker) for { id := trie.StateTrieID(dl.root) exhausted, last, err := dl.generateRange(ctx, id, rawdb.SnapshotAccountPrefix, snapAccount, origin, accountRange, onAccount, types.FullAccountRLP) if err != nil { return err // The procedure it aborted, either by external signal or internal error. } origin = increaseKey(last) // Last step, cleanup the storages after the last account. // All the left storages should be treated as dangling. if origin == nil || exhausted { ctx.removeStorageLeft() break } accountRange = accountCheckRange } return nil } // generate is a background thread that iterates over the state and storage tries, // constructing the state snapshot. All the arguments are purely for statistics // gathering and logging, since the method surfs the blocks as they arrive, often // being restarted. func (dl *diskLayer) generate(stats *generatorStats) { var ( accMarker []byte abort chan *generatorStats ) if len(dl.genMarker) > 0 { // []byte{} is the start, use nil for that accMarker = dl.genMarker[:common.HashLength] } stats.Log("Resuming state snapshot generation", dl.root, dl.genMarker) // Initialize the global generator context. The snapshot iterators are // opened at the interrupted position because the assumption is held // that all the snapshot data are generated correctly before the marker. // Even if the snapshot data is updated during the interruption (before // or at the marker), the assumption is still held. // For the account or storage slot at the interruption, they will be // processed twice by the generator(they are already processed in the // last run) but it's fine. ctx := newGeneratorContext(stats, dl.diskdb, accMarker, dl.genMarker) defer ctx.close() if err := generateAccounts(ctx, dl, accMarker); err != nil { // Extract the received interruption signal if exists if aerr, ok := err.(*abortErr); ok { abort = aerr.abort } // Aborted by internal error, wait the signal if abort == nil { abort = <-dl.genAbort } abort <- stats return } // Snapshot fully generated, set the marker to nil. // Note even there is nothing to commit, persist the // generator anyway to mark the snapshot is complete. journalProgress(ctx.batch, nil, stats) if err := ctx.batch.Write(); err != nil { log.Error("Failed to flush batch", "err", err) abort = <-dl.genAbort abort <- stats return } ctx.batch.Reset() log.Info("Generated state snapshot", "accounts", stats.accounts, "slots", stats.slots, "storage", stats.storage, "dangling", stats.dangling, "elapsed", common.PrettyDuration(time.Since(stats.start))) dl.lock.Lock() dl.genMarker = nil close(dl.genPending) dl.lock.Unlock() // Someone will be looking for us, wait it out abort = <-dl.genAbort abort <- nil } // increaseKey increase the input key by one bit. Return nil if the entire // addition operation overflows. func increaseKey(key []byte) []byte { for i := len(key) - 1; i >= 0; i-- { key[i]++ if key[i] != 0x0 { return key } } return nil } // abortErr wraps an interruption signal received to represent the // generation is aborted by external processes. type abortErr struct { abort chan *generatorStats } func newAbortErr(abort chan *generatorStats) error { return &abortErr{abort: abort} } func (err *abortErr) Error() string { return "aborted" }