cerc-io/plugeth
12
0
Fork 1
Code Issues 4 Pull Requests Packages Projects Releases Wiki Activity

all: remove the dependency from trie to triedb (#28824)

Browse Source 
This change removes the dependency from trie package to triedb package.
This commit is contained in:
rjl493456442
2024-02-13 14:49:53 +01:00
committed by GitHub
parent 4c15d58007
commit fe91d476ba
67 changed files with 597 additions and 425 deletions
Download Patch File Download Diff File Expand all files Collapse all files
trie/committer.go
+3 -3
View File
@@ -154,12 +154,12 @@ func (c *committer) store(path []byte, n node) node {
return hash
}
// mptResolver the children resolver in merkle-patricia-tree.
type mptResolver struct{}
// MerkleResolver the children resolver in merkle-patricia-tree.
type MerkleResolver struct{}
// ForEach implements childResolver, decodes the provided node and
// traverses the children inside.
func (resolver mptResolver) ForEach(node []byte, onChild func(common.Hash)) {
func (resolver MerkleResolver) ForEach(node []byte, onChild func(common.Hash)) {
forGatherChildren(mustDecodeNodeUnsafe(nil, node), onChild)
}
trie/database.go
-315
View File
@@ -1,315 +0,0 @@
// Copyright 2022 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 trie
import (
"errors"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/triedb/hashdb"
"github.com/ethereum/go-ethereum/trie/triedb/pathdb"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// Config defines all necessary options for database.
type Config struct {
Preimages bool // Flag whether the preimage of node key is recorded
IsVerkle bool // Flag whether the db is holding a verkle tree
HashDB *hashdb.Config // Configs for hash-based scheme
PathDB *pathdb.Config // Configs for experimental path-based scheme
}
// HashDefaults represents a config for using hash-based scheme with
// default settings.
var HashDefaults = &Config{
Preimages: false,
HashDB: hashdb.Defaults,
}
// backend defines the methods needed to access/update trie nodes in different
// state scheme.
type backend interface {
// Scheme returns the identifier of used storage scheme.
Scheme() string
// Initialized returns an indicator if the state data is already initialized
// according to the state scheme.
Initialized(genesisRoot common.Hash) bool
// Size returns the current storage size of the diff layers on top of the
// disk layer and the storage size of the nodes cached in the disk layer.
//
// For hash scheme, there is no differentiation between diff layer nodes
// and dirty disk layer nodes, so both are merged into the second return.
Size() (common.StorageSize, common.StorageSize)
// Update performs a state transition by committing dirty nodes contained
// in the given set in order to update state from the specified parent to
// the specified root.
//
// The passed in maps(nodes, states) will be retained to avoid copying
// everything. Therefore, these maps must not be changed afterwards.
Update(root common.Hash, parent common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error
// Commit writes all relevant trie nodes belonging to the specified state
// to disk. Report specifies whether logs will be displayed in info level.
Commit(root common.Hash, report bool) error
// Close closes the trie database backend and releases all held resources.
Close() error
}
// Database is the wrapper of the underlying backend which is shared by different
// types of node backend as an entrypoint. It's responsible for all interactions
// relevant with trie nodes and node preimages.
type Database struct {
config *Config // Configuration for trie database
diskdb ethdb.Database // Persistent database to store the snapshot
preimages *preimageStore // The store for caching preimages
backend backend // The backend for managing trie nodes
}
// NewDatabase initializes the trie database with default settings, note
// the legacy hash-based scheme is used by default.
func NewDatabase(diskdb ethdb.Database, config *Config) *Database {
// Sanitize the config and use the default one if it's not specified.
if config == nil {
config = HashDefaults
}
var preimages *preimageStore
if config.Preimages {
preimages = newPreimageStore(diskdb)
}
db := &Database{
config: config,
diskdb: diskdb,
preimages: preimages,
}
if config.HashDB != nil && config.PathDB != nil {
log.Crit("Both 'hash' and 'path' mode are configured")
}
if config.PathDB != nil {
db.backend = pathdb.New(diskdb, config.PathDB)
} else {
db.backend = hashdb.New(diskdb, config.HashDB, mptResolver{})
}
return db
}
// Reader returns a reader for accessing all trie nodes with provided state root.
// An error will be returned if the requested state is not available.
func (db *Database) Reader(blockRoot common.Hash) (Reader, error) {
switch b := db.backend.(type) {
case *hashdb.Database:
return b.Reader(blockRoot)
case *pathdb.Database:
return b.Reader(blockRoot)
}
return nil, errors.New("unknown backend")
}
// Update performs a state transition by committing dirty nodes contained in the
// given set in order to update state from the specified parent to the specified
// root. The held pre-images accumulated up to this point will be flushed in case
// the size exceeds the threshold.
//
// The passed in maps(nodes, states) will be retained to avoid copying everything.
// Therefore, these maps must not be changed afterwards.
func (db *Database) Update(root common.Hash, parent common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
if db.preimages != nil {
db.preimages.commit(false)
}
return db.backend.Update(root, parent, block, nodes, states)
}
// Commit iterates over all the children of a particular node, writes them out
// to disk. As a side effect, all pre-images accumulated up to this point are
// also written.
func (db *Database) Commit(root common.Hash, report bool) error {
if db.preimages != nil {
db.preimages.commit(true)
}
return db.backend.Commit(root, report)
}
// Size returns the storage size of diff layer nodes above the persistent disk
// layer, the dirty nodes buffered within the disk layer, and the size of cached
// preimages.
func (db *Database) Size() (common.StorageSize, common.StorageSize, common.StorageSize) {
var (
diffs, nodes common.StorageSize
preimages common.StorageSize
)
diffs, nodes = db.backend.Size()
if db.preimages != nil {
preimages = db.preimages.size()
}
return diffs, nodes, preimages
}
// Initialized returns an indicator if the state data is already initialized
// according to the state scheme.
func (db *Database) Initialized(genesisRoot common.Hash) bool {
return db.backend.Initialized(genesisRoot)
}
// Scheme returns the node scheme used in the database.
func (db *Database) Scheme() string {
return db.backend.Scheme()
}
// Close flushes the dangling preimages to disk and closes the trie database.
// It is meant to be called when closing the blockchain object, so that all
// resources held can be released correctly.
func (db *Database) Close() error {
db.WritePreimages()
return db.backend.Close()
}
// WritePreimages flushes all accumulated preimages to disk forcibly.
func (db *Database) WritePreimages() {
if db.preimages != nil {
db.preimages.commit(true)
}
}
// Preimage retrieves a cached trie node pre-image from memory. If it cannot be
// found cached, the method queries the persistent database for the content.
func (db *Database) Preimage(hash common.Hash) []byte {
if db.preimages == nil {
return nil
}
return db.preimages.preimage(hash)
}
// Cap iteratively flushes old but still referenced trie nodes until the total
// memory usage goes below the given threshold. The held pre-images accumulated
// up to this point will be flushed in case the size exceeds the threshold.
//
// It's only supported by hash-based database and will return an error for others.
func (db *Database) Cap(limit common.StorageSize) error {
hdb, ok := db.backend.(*hashdb.Database)
if !ok {
return errors.New("not supported")
}
if db.preimages != nil {
db.preimages.commit(false)
}
return hdb.Cap(limit)
}
// Reference adds a new reference from a parent node to a child node. This function
// is used to add reference between internal trie node and external node(e.g. storage
// trie root), all internal trie nodes are referenced together by database itself.
//
// It's only supported by hash-based database and will return an error for others.
func (db *Database) Reference(root common.Hash, parent common.Hash) error {
hdb, ok := db.backend.(*hashdb.Database)
if !ok {
return errors.New("not supported")
}
hdb.Reference(root, parent)
return nil
}
// Dereference removes an existing reference from a root node. It's only
// supported by hash-based database and will return an error for others.
func (db *Database) Dereference(root common.Hash) error {
hdb, ok := db.backend.(*hashdb.Database)
if !ok {
return errors.New("not supported")
}
hdb.Dereference(root)
return nil
}
// Recover rollbacks the database to a specified historical point. The state is
// supported as the rollback destination only if it's canonical state and the
// corresponding trie histories are existent. It's only supported by path-based
// database and will return an error for others.
func (db *Database) Recover(target common.Hash) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Recover(target, &trieLoader{db: db})
}
// Recoverable returns the indicator if the specified state is enabled to be
// recovered. It's only supported by path-based database and will return an
// error for others.
func (db *Database) Recoverable(root common.Hash) (bool, error) {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return false, errors.New("not supported")
}
return pdb.Recoverable(root), nil
}
// Disable deactivates the database and invalidates all available state layers
// as stale to prevent access to the persistent state, which is in the syncing
// stage.
//
// It's only supported by path-based database and will return an error for others.
func (db *Database) Disable() error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Disable()
}
// Enable activates database and resets the state tree with the provided persistent
// state root once the state sync is finished.
func (db *Database) Enable(root common.Hash) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Enable(root)
}
// Journal commits an entire diff hierarchy to disk into a single journal entry.
// This is meant to be used during shutdown to persist the snapshot without
// flattening everything down (bad for reorgs). It's only supported by path-based
// database and will return an error for others.
func (db *Database) Journal(root common.Hash) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Journal(root)
}
// SetBufferSize sets the node buffer size to the provided value(in bytes).
// It's only supported by path-based database and will return an error for
// others.
func (db *Database) SetBufferSize(size int) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.SetBufferSize(size)
}
// IsVerkle returns the indicator if the database is holding a verkle tree.
func (db *Database) IsVerkle() bool {
return db.config.IsVerkle
}
trie/database_test.go
+128 -16
View File
@@ -17,24 +17,136 @@
package trie
import (
"github.com/ethereum/go-ethereum/common"
"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/trie/triedb/hashdb"
"github.com/ethereum/go-ethereum/trie/triedb/pathdb"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/triedb/database"
)
// newTestDatabase initializes the trie database with specified scheme.
func newTestDatabase(diskdb ethdb.Database, scheme string) *Database {
config := &Config{Preimages: false}
if scheme == rawdb.HashScheme {
config.HashDB = &hashdb.Config{
CleanCacheSize: 0,
} // disable clean cache
} else {
config.PathDB = &pathdb.Config{
CleanCacheSize: 0,
DirtyCacheSize: 0,
} // disable clean/dirty cache
}
return NewDatabase(diskdb, config)
// testReader implements database.Reader interface, providing function to
// access trie nodes.
type testReader struct {
db ethdb.Database
scheme string
nodes []*trienode.MergedNodeSet // sorted from new to old
}
// Node implements database.Reader interface, retrieving trie node with
// all available cached layers.
func (r *testReader) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
// Check the node presence with the cached layer, from latest to oldest.
for _, nodes := range r.nodes {
if _, ok := nodes.Sets[owner]; !ok {
continue
}
n, ok := nodes.Sets[owner].Nodes[string(path)]
if !ok {
continue
}
if n.IsDeleted() || n.Hash != hash {
return nil, &MissingNodeError{Owner: owner, Path: path, NodeHash: hash}
}
return n.Blob, nil
}
// Check the node presence in database.
return rawdb.ReadTrieNode(r.db, owner, path, hash, r.scheme), nil
}
// testDb implements database.Database interface, using for testing purpose.
type testDb struct {
disk ethdb.Database
root common.Hash
scheme string
nodes map[common.Hash]*trienode.MergedNodeSet
parents map[common.Hash]common.Hash
}
func newTestDatabase(diskdb ethdb.Database, scheme string) *testDb {
return &testDb{
disk: diskdb,
root: types.EmptyRootHash,
scheme: scheme,
nodes: make(map[common.Hash]*trienode.MergedNodeSet),
parents: make(map[common.Hash]common.Hash),
}
}
func (db *testDb) Reader(stateRoot common.Hash) (database.Reader, error) {
nodes, _ := db.dirties(stateRoot, true)
return &testReader{db: db.disk, scheme: db.scheme, nodes: nodes}, nil
}
func (db *testDb) Preimage(hash common.Hash) []byte {
return rawdb.ReadPreimage(db.disk, hash)
}
func (db *testDb) InsertPreimage(preimages map[common.Hash][]byte) {
rawdb.WritePreimages(db.disk, preimages)
}
func (db *testDb) Scheme() string { return db.scheme }
func (db *testDb) Update(root common.Hash, parent common.Hash, nodes *trienode.MergedNodeSet) error {
if root == parent {
return nil
}
if _, ok := db.nodes[root]; ok {
return nil
}
db.parents[root] = parent
db.nodes[root] = nodes
return nil
}
func (db *testDb) dirties(root common.Hash, topToBottom bool) ([]*trienode.MergedNodeSet, []common.Hash) {
var (
pending []*trienode.MergedNodeSet
roots []common.Hash
)
for {
if root == db.root {
break
}
nodes, ok := db.nodes[root]
if !ok {
break
}
if topToBottom {
pending = append(pending, nodes)
roots = append(roots, root)
} else {
pending = append([]*trienode.MergedNodeSet{nodes}, pending...)
roots = append([]common.Hash{root}, roots...)
}
root = db.parents[root]
}
return pending, roots
}
func (db *testDb) Commit(root common.Hash) error {
if root == db.root {
return nil
}
pending, roots := db.dirties(root, false)
for i, nodes := range pending {
for owner, set := range nodes.Sets {
if owner == (common.Hash{}) {
continue
}
set.ForEachWithOrder(func(path string, n *trienode.Node) {
rawdb.WriteTrieNode(db.disk, owner, []byte(path), n.Hash, n.Blob, db.scheme)
})
}
nodes.Sets[common.Hash{}].ForEachWithOrder(func(path string, n *trienode.Node) {
rawdb.WriteTrieNode(db.disk, common.Hash{}, []byte(path), n.Hash, n.Blob, db.scheme)
})
db.root = roots[i]
}
for _, root := range roots {
delete(db.nodes, root)
delete(db.parents, root)
}
return nil
}
trie/iterator_test.go
+21 -20
View File
@@ -30,7 +30,7 @@ import (
)
func TestEmptyIterator(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
iter := trie.MustNodeIterator(nil)
seen := make(map[string]struct{})
@@ -43,7 +43,7 @@ func TestEmptyIterator(t *testing.T) {
}
func TestIterator(t *testing.T) {
db := NewDatabase(rawdb.NewMemoryDatabase(), nil)
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie := NewEmpty(db)
vals := []struct{ k, v string }{
{"do", "verb"},
@@ -60,7 +60,7 @@ func TestIterator(t *testing.T) {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
trie, _ = New(TrieID(root), db)
found := make(map[string]string)
@@ -86,7 +86,7 @@ func (k *kv) cmp(other *kv) int {
}
func TestIteratorLargeData(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
vals := make(map[string]*kv)
for i := byte(0); i < 255; i++ {
@@ -205,7 +205,7 @@ var testdata2 = []kvs{
}
func TestIteratorSeek(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
for _, val := range testdata1 {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
@@ -246,22 +246,22 @@ func checkIteratorOrder(want []kvs, it *Iterator) error {
}
func TestDifferenceIterator(t *testing.T) {
dba := NewDatabase(rawdb.NewMemoryDatabase(), nil)
dba := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
triea := NewEmpty(dba)
for _, val := range testdata1 {
triea.MustUpdate([]byte(val.k), []byte(val.v))
}
rootA, nodesA, _ := triea.Commit(false)
dba.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesA), nil)
dba.Update(rootA, types.EmptyRootHash, trienode.NewWithNodeSet(nodesA))
triea, _ = New(TrieID(rootA), dba)
dbb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
dbb := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trieb := NewEmpty(dbb)
for _, val := range testdata2 {
trieb.MustUpdate([]byte(val.k), []byte(val.v))
}
rootB, nodesB, _ := trieb.Commit(false)
dbb.Update(rootB, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesB), nil)
dbb.Update(rootB, types.EmptyRootHash, trienode.NewWithNodeSet(nodesB))
trieb, _ = New(TrieID(rootB), dbb)
found := make(map[string]string)
@@ -288,22 +288,22 @@ func TestDifferenceIterator(t *testing.T) {
}
func TestUnionIterator(t *testing.T) {
dba := NewDatabase(rawdb.NewMemoryDatabase(), nil)
dba := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
triea := NewEmpty(dba)
for _, val := range testdata1 {
triea.MustUpdate([]byte(val.k), []byte(val.v))
}
rootA, nodesA, _ := triea.Commit(false)
dba.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesA), nil)
dba.Update(rootA, types.EmptyRootHash, trienode.NewWithNodeSet(nodesA))
triea, _ = New(TrieID(rootA), dba)
dbb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
dbb := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trieb := NewEmpty(dbb)
for _, val := range testdata2 {
trieb.MustUpdate([]byte(val.k), []byte(val.v))
}
rootB, nodesB, _ := trieb.Commit(false)
dbb.Update(rootB, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesB), nil)
dbb.Update(rootB, types.EmptyRootHash, trienode.NewWithNodeSet(nodesB))
trieb, _ = New(TrieID(rootB), dbb)
di, _ := NewUnionIterator([]NodeIterator{triea.MustNodeIterator(nil), trieb.MustNodeIterator(nil)})
@@ -341,7 +341,8 @@ func TestUnionIterator(t *testing.T) {
}
func TestIteratorNoDups(t *testing.T) {
tr := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
tr := NewEmpty(db)
for _, val := range testdata1 {
tr.MustUpdate([]byte(val.k), []byte(val.v))
}
@@ -365,9 +366,9 @@ func testIteratorContinueAfterError(t *testing.T, memonly bool, scheme string) {
tr.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := tr.Commit(false)
tdb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
tdb.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
if !memonly {
tdb.Commit(root, false)
tdb.Commit(root)
}
tr, _ = New(TrieID(root), tdb)
wantNodeCount := checkIteratorNoDups(t, tr.MustNodeIterator(nil), nil)
@@ -481,9 +482,9 @@ func testIteratorContinueAfterSeekError(t *testing.T, memonly bool, scheme strin
break
}
}
triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
triedb.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
if !memonly {
triedb.Commit(root, false)
triedb.Commit(root)
}
var (
barNodeBlob []byte
@@ -555,8 +556,8 @@ func testIteratorNodeBlob(t *testing.T, scheme string) {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
triedb.Commit(root, false)
triedb.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
triedb.Commit(root)
var found = make(map[common.Hash][]byte)
trie, _ = New(TrieID(root), triedb)
trie/preimages.go
-95
View File
@@ -1,95 +0,0 @@
// Copyright 2022 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 trie
import (
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
)
// preimageStore is the store for caching preimages of node key.
type preimageStore struct {
lock sync.RWMutex
disk ethdb.KeyValueStore
preimages map[common.Hash][]byte // Preimages of nodes from the secure trie
preimagesSize common.StorageSize // Storage size of the preimages cache
}
// newPreimageStore initializes the store for caching preimages.
func newPreimageStore(disk ethdb.KeyValueStore) *preimageStore {
return &preimageStore{
disk: disk,
preimages: make(map[common.Hash][]byte),
}
}
// insertPreimage writes a new trie node pre-image to the memory database if it's
// yet unknown. The method will NOT make a copy of the slice, only use if the
// preimage will NOT be changed later on.
func (store *preimageStore) insertPreimage(preimages map[common.Hash][]byte) {
store.lock.Lock()
defer store.lock.Unlock()
for hash, preimage := range preimages {
if _, ok := store.preimages[hash]; ok {
continue
}
store.preimages[hash] = preimage
store.preimagesSize += common.StorageSize(common.HashLength + len(preimage))
}
}
// preimage retrieves a cached trie node pre-image from memory. If it cannot be
// found cached, the method queries the persistent database for the content.
func (store *preimageStore) preimage(hash common.Hash) []byte {
store.lock.RLock()
preimage := store.preimages[hash]
store.lock.RUnlock()
if preimage != nil {
return preimage
}
return rawdb.ReadPreimage(store.disk, hash)
}
// commit flushes the cached preimages into the disk.
func (store *preimageStore) commit(force bool) error {
store.lock.Lock()
defer store.lock.Unlock()
if store.preimagesSize <= 4*1024*1024 && !force {
return nil
}
batch := store.disk.NewBatch()
rawdb.WritePreimages(batch, store.preimages)
if err := batch.Write(); err != nil {
return err
}
store.preimages, store.preimagesSize = make(map[common.Hash][]byte), 0
return nil
}
// size returns the current storage size of accumulated preimages.
func (store *preimageStore) size() common.StorageSize {
store.lock.RLock()
defer store.lock.RUnlock()
return store.preimagesSize
}
trie/proof_test.go
+9 -9
View File
@@ -94,7 +94,7 @@ func TestProof(t *testing.T) {
}
func TestOneElementProof(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
updateString(trie, "k", "v")
for i, prover := range makeProvers(trie) {
proof := prover([]byte("k"))
@@ -145,7 +145,7 @@ func TestBadProof(t *testing.T) {
// Tests that missing keys can also be proven. The test explicitly uses a single
// entry trie and checks for missing keys both before and after the single entry.
func TestMissingKeyProof(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
updateString(trie, "k", "v")
for i, key := range []string{"a", "j", "l", "z"} {
@@ -343,7 +343,7 @@ func TestOneElementRangeProof(t *testing.T) {
}
// Test the mini trie with only a single element.
tinyTrie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
tinyTrie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
entry := &kv{randBytes(32), randBytes(20), false}
tinyTrie.MustUpdate(entry.k, entry.v)
@@ -414,7 +414,7 @@ func TestAllElementsProof(t *testing.T) {
// TestSingleSideRangeProof tests the range starts from zero.
func TestSingleSideRangeProof(t *testing.T) {
for i := 0; i < 64; i++ {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
var entries []*kv
for i := 0; i < 4096; i++ {
value := &kv{randBytes(32), randBytes(20), false}
@@ -520,7 +520,7 @@ func TestBadRangeProof(t *testing.T) {
// TestGappedRangeProof focuses on the small trie with embedded nodes.
// If the gapped node is embedded in the trie, it should be detected too.
func TestGappedRangeProof(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
var entries []*kv // Sorted entries
for i := byte(0); i < 10; i++ {
value := &kv{common.LeftPadBytes([]byte{i}, 32), []byte{i}, false}
@@ -592,7 +592,7 @@ func TestSameSideProofs(t *testing.T) {
}
func TestHasRightElement(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
var entries []*kv
for i := 0; i < 4096; i++ {
value := &kv{randBytes(32), randBytes(20), false}
@@ -934,7 +934,7 @@ func benchmarkVerifyRangeNoProof(b *testing.B, size int) {
}
func randomTrie(n int) (*Trie, map[string]*kv) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
vals := make(map[string]*kv)
for i := byte(0); i < 100; i++ {
value := &kv{common.LeftPadBytes([]byte{i}, 32), []byte{i}, false}
@@ -953,7 +953,7 @@ func randomTrie(n int) (*Trie, map[string]*kv) {
}
func nonRandomTrie(n int) (*Trie, map[string]*kv) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
vals := make(map[string]*kv)
max := uint64(0xffffffffffffffff)
for i := uint64(0); i < uint64(n); i++ {
@@ -978,7 +978,7 @@ func TestRangeProofKeysWithSharedPrefix(t *testing.T) {
common.Hex2Bytes("02"),
common.Hex2Bytes("03"),
}
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
for i, key := range keys {
trie.MustUpdate(key, vals[i])
}
trie/secure_trie.go
+11 -15
View File
@@ -21,6 +21,7 @@ import (
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/triedb/database"
)
// SecureTrie is the old name of StateTrie.
@@ -29,7 +30,7 @@ type SecureTrie = StateTrie
// NewSecure creates a new StateTrie.
// Deprecated: use NewStateTrie.
func NewSecure(stateRoot common.Hash, owner common.Hash, root common.Hash, db *Database) (*SecureTrie, error) {
func NewSecure(stateRoot common.Hash, owner common.Hash, root common.Hash, db database.Database) (*SecureTrie, error) {
id := &ID{
StateRoot: stateRoot,
Owner: owner,
@@ -50,7 +51,7 @@ func NewSecure(stateRoot common.Hash, owner common.Hash, root common.Hash, db *D
// StateTrie is not safe for concurrent use.
type StateTrie struct {
trie Trie
preimages *preimageStore
db database.Database
hashKeyBuf [common.HashLength]byte
secKeyCache map[string][]byte
secKeyCacheOwner *StateTrie // Pointer to self, replace the key cache on mismatch
@@ -61,7 +62,7 @@ type StateTrie struct {
// If root is the zero hash or the sha3 hash of an empty string, the
// trie is initially empty. Otherwise, New will panic if db is nil
// and returns MissingNodeError if the root node cannot be found.
func NewStateTrie(id *ID, db *Database) (*StateTrie, error) {
func NewStateTrie(id *ID, db database.Database) (*StateTrie, error) {
if db == nil {
panic("trie.NewStateTrie called without a database")
}
@@ -69,7 +70,7 @@ func NewStateTrie(id *ID, db *Database) (*StateTrie, error) {
if err != nil {
return nil, err
}
return &StateTrie{trie: *trie, preimages: db.preimages}, nil
return &StateTrie{trie: *trie, db: db}, nil
}
// MustGet returns the value for key stored in the trie.
@@ -210,10 +211,7 @@ func (t *StateTrie) GetKey(shaKey []byte) []byte {
if key, ok := t.getSecKeyCache()[string(shaKey)]; ok {
return key
}
if t.preimages == nil {
return nil
}
return t.preimages.preimage(common.BytesToHash(shaKey))
return t.db.Preimage(common.BytesToHash(shaKey))
}
// Commit collects all dirty nodes in the trie and replaces them with the
@@ -226,13 +224,11 @@ func (t *StateTrie) GetKey(shaKey []byte) []byte {
func (t *StateTrie) Commit(collectLeaf bool) (common.Hash, *trienode.NodeSet, error) {
// Write all the pre-images to the actual disk database
if len(t.getSecKeyCache()) > 0 {
if t.preimages != nil {
preimages := make(map[common.Hash][]byte)
for hk, key := range t.secKeyCache {
preimages[common.BytesToHash([]byte(hk))] = key
}
t.preimages.insertPreimage(preimages)
preimages := make(map[common.Hash][]byte)
for hk, key := range t.secKeyCache {
preimages[common.BytesToHash([]byte(hk))] = key
}
t.db.InsertPreimage(preimages)
t.secKeyCache = make(map[string][]byte)
}
// Commit the trie and return its modified nodeset.
@@ -249,7 +245,7 @@ func (t *StateTrie) Hash() common.Hash {
func (t *StateTrie) Copy() *StateTrie {
return &StateTrie{
trie: *t.trie.Copy(),
preimages: t.preimages,
db: t.db,
secKeyCache: t.secKeyCache,
}
}
trie/secure_trie_test.go
+4 -4
View File
@@ -31,14 +31,14 @@ import (
)
func newEmptySecure() *StateTrie {
trie, _ := NewStateTrie(TrieID(types.EmptyRootHash), NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie, _ := NewStateTrie(TrieID(types.EmptyRootHash), newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
return trie
}
// makeTestStateTrie creates a large enough secure trie for testing.
func makeTestStateTrie() (*Database, *StateTrie, map[string][]byte) {
func makeTestStateTrie() (*testDb, *StateTrie, map[string][]byte) {
// Create an empty trie
triedb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
triedb := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie, _ := NewStateTrie(TrieID(types.EmptyRootHash), triedb)
// Fill it with some arbitrary data
@@ -61,7 +61,7 @@ func makeTestStateTrie() (*Database, *StateTrie, map[string][]byte) {
}
}
root, nodes, _ := trie.Commit(false)
if err := triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
if err := triedb.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes)); err != nil {
panic(fmt.Errorf("failed to commit db %v", err))
}
// Re-create the trie based on the new state
trie/stacktrie_fuzzer_test.go
+4 -4
View File
@@ -42,10 +42,10 @@ func fuzz(data []byte, debugging bool) {
var (
input = bytes.NewReader(data)
spongeA = &spongeDb{sponge: sha3.NewLegacyKeccak256()}
dbA = NewDatabase(rawdb.NewDatabase(spongeA), nil)
dbA = newTestDatabase(rawdb.NewDatabase(spongeA), rawdb.HashScheme)
trieA = NewEmpty(dbA)
spongeB = &spongeDb{sponge: sha3.NewLegacyKeccak256()}
dbB = NewDatabase(rawdb.NewDatabase(spongeB), nil)
dbB = newTestDatabase(rawdb.NewDatabase(spongeB), rawdb.HashScheme)
options = NewStackTrieOptions().WithWriter(func(path []byte, hash common.Hash, blob []byte) {
rawdb.WriteTrieNode(spongeB, common.Hash{}, path, hash, blob, dbB.Scheme())
@@ -87,10 +87,10 @@ func fuzz(data []byte, debugging bool) {
panic(err)
}
if nodes != nil {
dbA.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
dbA.Update(rootA, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
}
// Flush memdb -> disk (sponge)
dbA.Commit(rootA, false)
dbA.Commit(rootA)
// Stacktrie requires sorted insertion
slices.SortFunc(vals, (*kv).cmp)
trie/stacktrie_test.go
+5 -5
View File
@@ -223,7 +223,7 @@ func TestStackTrieInsertAndHash(t *testing.T) {
func TestSizeBug(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
nt := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
leaf := common.FromHex("290decd9548b62a8d60345a988386fc84ba6bc95484008f6362f93160ef3e563")
value := common.FromHex("94cf40d0d2b44f2b66e07cace1372ca42b73cf21a3")
@@ -238,7 +238,7 @@ func TestSizeBug(t *testing.T) {
func TestEmptyBug(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
nt := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
//leaf := common.FromHex("290decd9548b62a8d60345a988386fc84ba6bc95484008f6362f93160ef3e563")
//value := common.FromHex("94cf40d0d2b44f2b66e07cace1372ca42b73cf21a3")
@@ -264,7 +264,7 @@ func TestEmptyBug(t *testing.T) {
func TestValLength56(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
nt := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
//leaf := common.FromHex("290decd9548b62a8d60345a988386fc84ba6bc95484008f6362f93160ef3e563")
//value := common.FromHex("94cf40d0d2b44f2b66e07cace1372ca42b73cf21a3")
@@ -289,7 +289,7 @@ func TestValLength56(t *testing.T) {
// which causes a lot of node-within-node. This case was found via fuzzing.
func TestUpdateSmallNodes(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
nt := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
kvs := []struct {
K string
V string
@@ -317,7 +317,7 @@ func TestUpdateSmallNodes(t *testing.T) {
func TestUpdateVariableKeys(t *testing.T) {
t.SkipNow()
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
nt := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
kvs := []struct {
K string
V string
trie/sync_test.go
+22 -22
View File
@@ -32,7 +32,7 @@ import (
)
// makeTestTrie create a sample test trie to test node-wise reconstruction.
func makeTestTrie(scheme string) (ethdb.Database, *Database, *StateTrie, map[string][]byte) {
func makeTestTrie(scheme string) (ethdb.Database, *testDb, *StateTrie, map[string][]byte) {
// Create an empty trie
db := rawdb.NewMemoryDatabase()
triedb := newTestDatabase(db, scheme)
@@ -58,10 +58,10 @@ func makeTestTrie(scheme string) (ethdb.Database, *Database, *StateTrie, map[str
}
}
root, nodes, _ := trie.Commit(false)
if err := triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
if err := triedb.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes)); err != nil {
panic(fmt.Errorf("failed to commit db %v", err))
}
if err := triedb.Commit(root, false); err != nil {
if err := triedb.Commit(root); err != nil {
panic(err)
}
// Re-create the trie based on the new state
@@ -143,7 +143,7 @@ func TestEmptySync(t *testing.T) {
emptyD, _ := New(TrieID(types.EmptyRootHash), dbD)
for i, trie := range []*Trie{emptyA, emptyB, emptyC, emptyD} {
sync := NewSync(trie.Hash(), memorydb.New(), nil, []*Database{dbA, dbB, dbC, dbD}[i].Scheme())
sync := NewSync(trie.Hash(), memorydb.New(), nil, []*testDb{dbA, dbB, dbC, dbD}[i].Scheme())
if paths, nodes, codes := sync.Missing(1); len(paths) != 0 || len(nodes) != 0 || len(codes) != 0 {
t.Errorf("test %d: content requested for empty trie: %v, %v, %v", i, paths, nodes, codes)
}
@@ -684,11 +684,11 @@ func testSyncOrdering(t *testing.T, scheme string) {
}
}
}
func syncWith(t *testing.T, root common.Hash, db ethdb.Database, srcDb *Database) {
func syncWith(t *testing.T, root common.Hash, db ethdb.Database, srcDb *testDb) {
syncWithHookWriter(t, root, db, srcDb, nil)
}
func syncWithHookWriter(t *testing.T, root common.Hash, db ethdb.Database, srcDb *Database, hookWriter ethdb.KeyValueWriter) {
func syncWithHookWriter(t *testing.T, root common.Hash, db ethdb.Database, srcDb *testDb, hookWriter ethdb.KeyValueWriter) {
// Create a destination trie and sync with the scheduler
sched := NewSync(root, db, nil, srcDb.Scheme())
@@ -771,10 +771,10 @@ func testSyncMovingTarget(t *testing.T, scheme string) {
diff[string(key)] = val
}
root, nodes, _ := srcTrie.Commit(false)
if err := srcDb.Update(root, preRoot, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
if err := srcDb.Update(root, preRoot, trienode.NewWithNodeSet(nodes)); err != nil {
panic(err)
}
if err := srcDb.Commit(root, false); err != nil {
if err := srcDb.Commit(root); err != nil {
panic(err)
}
preRoot = root
@@ -796,10 +796,10 @@ func testSyncMovingTarget(t *testing.T, scheme string) {
reverted[k] = val
}
root, nodes, _ = srcTrie.Commit(false)
if err := srcDb.Update(root, preRoot, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
if err := srcDb.Update(root, preRoot, trienode.NewWithNodeSet(nodes)); err != nil {
panic(err)
}
if err := srcDb.Commit(root, false); err != nil {
if err := srcDb.Commit(root); err != nil {
panic(err)
}
srcTrie, _ = NewStateTrie(TrieID(root), srcDb)
@@ -854,10 +854,10 @@ func testPivotMove(t *testing.T, scheme string, tiny bool) {
writeFn([]byte{0x13, 0x44}, nil, srcTrie, stateA)
rootA, nodesA, _ := srcTrie.Commit(false)
if err := srcTrieDB.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesA), nil); err != nil {
if err := srcTrieDB.Update(rootA, types.EmptyRootHash, trienode.NewWithNodeSet(nodesA)); err != nil {
panic(err)
}
if err := srcTrieDB.Commit(rootA, false); err != nil {
if err := srcTrieDB.Commit(rootA); err != nil {
panic(err)
}
// Create a destination trie and sync with the scheduler
@@ -873,10 +873,10 @@ func testPivotMove(t *testing.T, scheme string, tiny bool) {
writeFn([]byte{0x01, 0x24}, nil, srcTrie, stateB)
rootB, nodesB, _ := srcTrie.Commit(false)
if err := srcTrieDB.Update(rootB, rootA, 0, trienode.NewWithNodeSet(nodesB), nil); err != nil {
if err := srcTrieDB.Update(rootB, rootA, trienode.NewWithNodeSet(nodesB)); err != nil {
panic(err)
}
if err := srcTrieDB.Commit(rootB, false); err != nil {
if err := srcTrieDB.Commit(rootB); err != nil {
panic(err)
}
syncWith(t, rootB, destDisk, srcTrieDB)
@@ -891,10 +891,10 @@ func testPivotMove(t *testing.T, scheme string, tiny bool) {
writeFn([]byte{0x13, 0x44}, nil, srcTrie, stateC)
rootC, nodesC, _ := srcTrie.Commit(false)
if err := srcTrieDB.Update(rootC, rootB, 0, trienode.NewWithNodeSet(nodesC), nil); err != nil {
if err := srcTrieDB.Update(rootC, rootB, trienode.NewWithNodeSet(nodesC)); err != nil {
panic(err)
}
if err := srcTrieDB.Commit(rootC, false); err != nil {
if err := srcTrieDB.Commit(rootC); err != nil {
panic(err)
}
syncWith(t, rootC, destDisk, srcTrieDB)
@@ -960,10 +960,10 @@ func testSyncAbort(t *testing.T, scheme string) {
writeFn(key, val, srcTrie, stateA)
rootA, nodesA, _ := srcTrie.Commit(false)
if err := srcTrieDB.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesA), nil); err != nil {
if err := srcTrieDB.Update(rootA, types.EmptyRootHash, trienode.NewWithNodeSet(nodesA)); err != nil {
panic(err)
}
if err := srcTrieDB.Commit(rootA, false); err != nil {
if err := srcTrieDB.Commit(rootA); err != nil {
panic(err)
}
// Create a destination trie and sync with the scheduler
@@ -977,10 +977,10 @@ func testSyncAbort(t *testing.T, scheme string) {
deleteFn(key, srcTrie, stateB)
rootB, nodesB, _ := srcTrie.Commit(false)
if err := srcTrieDB.Update(rootB, rootA, 0, trienode.NewWithNodeSet(nodesB), nil); err != nil {
if err := srcTrieDB.Update(rootB, rootA, trienode.NewWithNodeSet(nodesB)); err != nil {
panic(err)
}
if err := srcTrieDB.Commit(rootB, false); err != nil {
if err := srcTrieDB.Commit(rootB); err != nil {
panic(err)
}
@@ -1004,10 +1004,10 @@ func testSyncAbort(t *testing.T, scheme string) {
writeFn(key, val, srcTrie, stateC)
rootC, nodesC, _ := srcTrie.Commit(false)
if err := srcTrieDB.Update(rootC, rootB, 0, trienode.NewWithNodeSet(nodesC), nil); err != nil {
if err := srcTrieDB.Update(rootC, rootB, trienode.NewWithNodeSet(nodesC)); err != nil {
panic(err)
}
if err := srcTrieDB.Commit(rootC, false); err != nil {
if err := srcTrieDB.Commit(rootC); err != nil {
panic(err)
}
syncWith(t, rootC, destDisk, srcTrieDB)
trie/tracer_test.go
+16 -15
View File
@@ -61,7 +61,7 @@ func TestTrieTracer(t *testing.T) {
// Tests if the trie diffs are tracked correctly. Tracer should capture
// all non-leaf dirty nodes, no matter the node is embedded or not.
func testTrieTracer(t *testing.T, vals []struct{ k, v string }) {
db := NewDatabase(rawdb.NewMemoryDatabase(), nil)
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie := NewEmpty(db)
// Determine all new nodes are tracked
@@ -71,7 +71,7 @@ func testTrieTracer(t *testing.T, vals []struct{ k, v string }) {
insertSet := copySet(trie.tracer.inserts) // copy before commit
deleteSet := copySet(trie.tracer.deletes) // copy before commit
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
seen := setKeys(iterNodes(db, root))
if !compareSet(insertSet, seen) {
@@ -104,7 +104,8 @@ func TestTrieTracerNoop(t *testing.T) {
}
func testTrieTracerNoop(t *testing.T, vals []struct{ k, v string }) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie := NewEmpty(db)
for _, val := range vals {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
@@ -128,7 +129,7 @@ func TestAccessList(t *testing.T) {
func testAccessList(t *testing.T, vals []struct{ k, v string }) {
var (
db = NewDatabase(rawdb.NewMemoryDatabase(), nil)
db = newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie = NewEmpty(db)
orig = trie.Copy()
)
@@ -137,7 +138,7 @@ func testAccessList(t *testing.T, vals []struct{ k, v string }) {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
@@ -152,7 +153,7 @@ func testAccessList(t *testing.T, vals []struct{ k, v string }) {
trie.MustUpdate([]byte(val.k), randBytes(32))
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, parent, trienode.NewWithNodeSet(nodes))
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
@@ -170,7 +171,7 @@ func testAccessList(t *testing.T, vals []struct{ k, v string }) {
trie.MustUpdate(key, randBytes(32))
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, parent, trienode.NewWithNodeSet(nodes))
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
@@ -185,7 +186,7 @@ func testAccessList(t *testing.T, vals []struct{ k, v string }) {
trie.MustUpdate([]byte(key), nil)
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, parent, trienode.NewWithNodeSet(nodes))
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
@@ -200,7 +201,7 @@ func testAccessList(t *testing.T, vals []struct{ k, v string }) {
trie.MustUpdate([]byte(val.k), nil)
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, parent, trienode.NewWithNodeSet(nodes))
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
@@ -211,7 +212,7 @@ func testAccessList(t *testing.T, vals []struct{ k, v string }) {
// Tests origin values won't be tracked in Iterator or Prover
func TestAccessListLeak(t *testing.T) {
var (
db = NewDatabase(rawdb.NewMemoryDatabase(), nil)
db = newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie = NewEmpty(db)
)
// Create trie from scratch
@@ -219,7 +220,7 @@ func TestAccessListLeak(t *testing.T) {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
var cases = []struct {
op func(tr *Trie)
@@ -262,14 +263,14 @@ func TestAccessListLeak(t *testing.T) {
// in its parent due to the smaller size of the original tree node.
func TestTinyTree(t *testing.T) {
var (
db = NewDatabase(rawdb.NewMemoryDatabase(), nil)
db = newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie = NewEmpty(db)
)
for _, val := range tiny {
trie.MustUpdate([]byte(val.k), randBytes(32))
}
root, set, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(set), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(set))
parent := root
trie, _ = New(TrieID(root), db)
@@ -278,7 +279,7 @@ func TestTinyTree(t *testing.T) {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, set, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(set), nil)
db.Update(root, parent, trienode.NewWithNodeSet(set))
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, set); err != nil {
@@ -312,7 +313,7 @@ func forNodes(tr *Trie) map[string][]byte {
return nodes
}
func iterNodes(db *Database, root common.Hash) map[string][]byte {
func iterNodes(db *testDb, root common.Hash) map[string][]byte {
tr, _ := New(TrieID(root), db)
return forNodes(tr)
}
trie/trie.go
+3 -2
View File
@@ -26,6 +26,7 @@ import (
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/triedb/database"
)
// Trie is a Merkle Patricia Trie. Use New to create a trie that sits on
@@ -79,7 +80,7 @@ func (t *Trie) Copy() *Trie {
// zero hash or the sha3 hash of an empty string, then trie is initially
// empty, otherwise, the root node must be present in database or returns
// a MissingNodeError if not.
func New(id *ID, db *Database) (*Trie, error) {
func New(id *ID, db database.Database) (*Trie, error) {
reader, err := newTrieReader(id.StateRoot, id.Owner, db)
if err != nil {
return nil, err
@@ -100,7 +101,7 @@ func New(id *ID, db *Database) (*Trie, error) {
}
// NewEmpty is a shortcut to create empty tree. It's mostly used in tests.
func NewEmpty(db *Database) *Trie {
func NewEmpty(db database.Database) *Trie {
tr, _ := New(TrieID(types.EmptyRootHash), db)
return tr
}
trie/trie_reader.go
+13 -20
View File
@@ -21,31 +21,19 @@ import (
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/triestate"
"github.com/ethereum/go-ethereum/triedb/database"
)
// Reader wraps the Node method of a backing trie store.
type Reader interface {
// Node retrieves the trie node blob with the provided trie identifier, node path and
// the corresponding node hash. No error will be returned if the node is not found.
//
// When looking up nodes in the account trie, 'owner' is the zero hash. For contract
// storage trie nodes, 'owner' is the hash of the account address that containing the
// storage.
//
// TODO(rjl493456442): remove the 'hash' parameter, it's redundant in PBSS.
Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error)
}
// trieReader is a wrapper of the underlying node reader. It's not safe
// for concurrent usage.
type trieReader struct {
owner common.Hash
reader Reader
reader database.Reader
banned map[string]struct{} // Marker to prevent node from being accessed, for tests
}
// newTrieReader initializes the trie reader with the given node reader.
func newTrieReader(stateRoot, owner common.Hash, db *Database) (*trieReader, error) {
func newTrieReader(stateRoot, owner common.Hash, db database.Database) (*trieReader, error) {
if stateRoot == (common.Hash{}) || stateRoot == types.EmptyRootHash {
if stateRoot == (common.Hash{}) {
log.Error("Zero state root hash!")
@@ -85,17 +73,22 @@ func (r *trieReader) node(path []byte, hash common.Hash) ([]byte, error) {
return blob, nil
}
// trieLoader implements triestate.TrieLoader for constructing tries.
type trieLoader struct {
db *Database
// MerkleLoader implements triestate.TrieLoader for constructing tries.
type MerkleLoader struct {
db database.Database
}
// NewMerkleLoader creates the merkle trie loader.
func NewMerkleLoader(db database.Database) *MerkleLoader {
return &MerkleLoader{db: db}
}
// OpenTrie opens the main account trie.
func (l *trieLoader) OpenTrie(root common.Hash) (triestate.Trie, error) {
func (l *MerkleLoader) OpenTrie(root common.Hash) (triestate.Trie, error) {
return New(TrieID(root), l.db)
}
// OpenStorageTrie opens the storage trie of an account.
func (l *trieLoader) OpenStorageTrie(stateRoot common.Hash, addrHash, root common.Hash) (triestate.Trie, error) {
func (l *MerkleLoader) OpenStorageTrie(stateRoot common.Hash, addrHash, root common.Hash) (triestate.Trie, error) {
return New(StorageTrieID(stateRoot, addrHash, root), l.db)
}
trie/trie_test.go
+95 -105
View File
@@ -25,6 +25,7 @@ import (
"io"
"math/rand"
"reflect"
"sort"
"testing"
"testing/quick"
@@ -46,7 +47,7 @@ func init() {
}
func TestEmptyTrie(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
res := trie.Hash()
exp := types.EmptyRootHash
if res != exp {
@@ -55,7 +56,7 @@ func TestEmptyTrie(t *testing.T) {
}
func TestNull(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
key := make([]byte, 32)
value := []byte("test")
trie.MustUpdate(key, value)
@@ -95,10 +96,10 @@ func testMissingNode(t *testing.T, memonly bool, scheme string) {
updateString(trie, "120000", "qwerqwerqwerqwerqwerqwerqwerqwer")
updateString(trie, "123456", "asdfasdfasdfasdfasdfasdfasdfasdf")
root, nodes, _ := trie.Commit(false)
triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
triedb.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
if !memonly {
triedb.Commit(root, false)
triedb.Commit(root)
}
trie, _ = New(TrieID(root), triedb)
@@ -167,7 +168,7 @@ func testMissingNode(t *testing.T, memonly bool, scheme string) {
}
func TestInsert(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
updateString(trie, "doe", "reindeer")
updateString(trie, "dog", "puppy")
@@ -179,7 +180,7 @@ func TestInsert(t *testing.T) {
t.Errorf("case 1: exp %x got %x", exp, root)
}
trie = NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie = NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
updateString(trie, "A", "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa")
exp = common.HexToHash("d23786fb4a010da3ce639d66d5e904a11dbc02746d1ce25029e53290cabf28ab")
@@ -190,7 +191,7 @@ func TestInsert(t *testing.T) {
}
func TestGet(t *testing.T) {
db := NewDatabase(rawdb.NewMemoryDatabase(), nil)
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie := NewEmpty(db)
updateString(trie, "doe", "reindeer")
updateString(trie, "dog", "puppy")
@@ -209,13 +210,14 @@ func TestGet(t *testing.T) {
return
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
trie, _ = New(TrieID(root), db)
}
}
func TestDelete(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie := NewEmpty(db)
vals := []struct{ k, v string }{
{"do", "verb"},
{"ether", "wookiedoo"},
@@ -242,7 +244,7 @@ func TestDelete(t *testing.T) {
}
func TestEmptyValues(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
vals := []struct{ k, v string }{
{"do", "verb"},
@@ -266,7 +268,7 @@ func TestEmptyValues(t *testing.T) {
}
func TestReplication(t *testing.T) {
db := NewDatabase(rawdb.NewMemoryDatabase(), nil)
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie := NewEmpty(db)
vals := []struct{ k, v string }{
{"do", "verb"},
@@ -281,7 +283,7 @@ func TestReplication(t *testing.T) {
updateString(trie, val.k, val.v)
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
// create a new trie on top of the database and check that lookups work.
trie2, err := New(TrieID(root), db)
@@ -300,7 +302,7 @@ func TestReplication(t *testing.T) {
// recreate the trie after commit
if nodes != nil {
db.Update(hash, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(hash, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
}
trie2, err = New(TrieID(hash), db)
if err != nil {
@@ -327,7 +329,7 @@ func TestReplication(t *testing.T) {
}
func TestLargeValue(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
trie.MustUpdate([]byte("key1"), []byte{99, 99, 99, 99})
trie.MustUpdate([]byte("key2"), bytes.Repeat([]byte{1}, 32))
trie.Hash()
@@ -531,7 +533,7 @@ func runRandTest(rt randTest) error {
case opCommit:
root, nodes, _ := tr.Commit(true)
if nodes != nil {
triedb.Update(root, origin, 0, trienode.NewWithNodeSet(nodes), nil)
triedb.Update(root, origin, trienode.NewWithNodeSet(nodes))
}
newtr, err := New(TrieID(root), triedb)
if err != nil {
@@ -632,7 +634,7 @@ func BenchmarkUpdateLE(b *testing.B) { benchUpdate(b, binary.LittleEndian) }
const benchElemCount = 20000
func benchGet(b *testing.B) {
triedb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
triedb := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme)
trie := NewEmpty(triedb)
k := make([]byte, 32)
for i := 0; i < benchElemCount; i++ {
@@ -651,7 +653,7 @@ func benchGet(b *testing.B) {
}
func benchUpdate(b *testing.B, e binary.ByteOrder) *Trie {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
k := make([]byte, 32)
b.ReportAllocs()
for i := 0; i < b.N; i++ {
@@ -683,7 +685,7 @@ func BenchmarkHash(b *testing.B) {
// entries, then adding N more.
addresses, accounts := makeAccounts(2 * b.N)
// Insert the accounts into the trie and hash it
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
i := 0
for ; i < len(addresses)/2; i++ {
trie.MustUpdate(crypto.Keccak256(addresses[i][:]), accounts[i])
@@ -714,7 +716,7 @@ func BenchmarkCommitAfterHash(b *testing.B) {
func benchmarkCommitAfterHash(b *testing.B, collectLeaf bool) {
// Make the random benchmark deterministic
addresses, accounts := makeAccounts(b.N)
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
for i := 0; i < len(addresses); i++ {
trie.MustUpdate(crypto.Keccak256(addresses[i][:]), accounts[i])
}
@@ -728,7 +730,7 @@ func benchmarkCommitAfterHash(b *testing.B, collectLeaf bool) {
func TestTinyTrie(t *testing.T) {
// Create a realistic account trie to hash
_, accounts := makeAccounts(5)
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
trie.MustUpdate(common.Hex2Bytes("0000000000000000000000000000000000000000000000000000000000001337"), accounts[3])
if exp, root := common.HexToHash("8c6a85a4d9fda98feff88450299e574e5378e32391f75a055d470ac0653f1005"), trie.Hash(); exp != root {
t.Errorf("1: got %x, exp %x", root, exp)
@@ -741,7 +743,7 @@ func TestTinyTrie(t *testing.T) {
if exp, root := common.HexToHash("0608c1d1dc3905fa22204c7a0e43644831c3b6d3def0f274be623a948197e64a"), trie.Hash(); exp != root {
t.Errorf("3: got %x, exp %x", root, exp)
}
checktr := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
checktr := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
it := NewIterator(trie.MustNodeIterator(nil))
for it.Next() {
checktr.MustUpdate(it.Key, it.Value)
@@ -754,7 +756,7 @@ func TestTinyTrie(t *testing.T) {
func TestCommitAfterHash(t *testing.T) {
// Create a realistic account trie to hash
addresses, accounts := makeAccounts(1000)
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
for i := 0; i < len(addresses); i++ {
trie.MustUpdate(crypto.Keccak256(addresses[i][:]), accounts[i])
}
@@ -808,6 +810,8 @@ type spongeDb struct {
sponge hash.Hash
id string
journal []string
keys []string
values map[string]string
}
func (s *spongeDb) Has(key []byte) (bool, error) { panic("implement me") }
@@ -831,12 +835,27 @@ func (s *spongeDb) Put(key []byte, value []byte) error {
valbrief = valbrief[:8]
}
s.journal = append(s.journal, fmt.Sprintf("%v: PUT([%x...], [%d bytes] %x...)\n", s.id, keybrief, len(value), valbrief))
s.sponge.Write(key)
s.sponge.Write(value)
if s.values == nil {
s.sponge.Write(key)
s.sponge.Write(value)
} else {
s.keys = append(s.keys, string(key))
s.values[string(key)] = string(value)
}
return nil
}
func (s *spongeDb) NewIterator(prefix []byte, start []byte) ethdb.Iterator { panic("implement me") }
func (s *spongeDb) Flush() {
// Bottom-up, the longest path first
sort.Sort(sort.Reverse(sort.StringSlice(s.keys)))
for _, key := range s.keys {
s.sponge.Write([]byte(key))
s.sponge.Write([]byte(s.values[key]))
}
}
// spongeBatch is a dummy batch which immediately writes to the underlying spongedb
type spongeBatch struct {
db *spongeDb
@@ -861,14 +880,14 @@ func TestCommitSequence(t *testing.T) {
count int
expWriteSeqHash []byte
}{
{20, common.FromHex("873c78df73d60e59d4a2bcf3716e8bfe14554549fea2fc147cb54129382a8066")},
{200, common.FromHex("ba03d891bb15408c940eea5ee3d54d419595102648d02774a0268d892add9c8e")},
{2000, common.FromHex("f7a184f20df01c94f09537401d11e68d97ad0c00115233107f51b9c287ce60c7")},
{20, common.FromHex("330b0afae2853d96b9f015791fbe0fb7f239bf65f335f16dfc04b76c7536276d")},
{200, common.FromHex("5162b3735c06b5d606b043a3ee8adbdbbb408543f4966bca9dcc63da82684eeb")},
{2000, common.FromHex("4574cd8e6b17f3fe8ad89140d1d0bf4f1bd7a87a8ac3fb623b33550544c77635")},
} {
addresses, accounts := makeAccounts(tc.count)
// This spongeDb is used to check the sequence of disk-db-writes
s := &spongeDb{sponge: sha3.NewLegacyKeccak256()}
db := NewDatabase(rawdb.NewDatabase(s), nil)
db := newTestDatabase(rawdb.NewDatabase(s), rawdb.HashScheme)
trie := NewEmpty(db)
// Fill the trie with elements
for i := 0; i < tc.count; i++ {
@@ -876,9 +895,9 @@ func TestCommitSequence(t *testing.T) {
}
// Flush trie -> database
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
// Flush memdb -> disk (sponge)
db.Commit(root, false)
db.Commit(root)
if got, exp := s.sponge.Sum(nil), tc.expWriteSeqHash; !bytes.Equal(got, exp) {
t.Errorf("test %d, disk write sequence wrong:\ngot %x exp %x\n", i, got, exp)
}
@@ -892,14 +911,14 @@ func TestCommitSequenceRandomBlobs(t *testing.T) {
count int
expWriteSeqHash []byte
}{
{20, common.FromHex("8e4a01548551d139fa9e833ebc4e66fc1ba40a4b9b7259d80db32cff7b64ebbc")},
{200, common.FromHex("6869b4e7b95f3097a19ddb30ff735f922b915314047e041614df06958fc50554")},
{2000, common.FromHex("444200e6f4e2df49f77752f629a96ccf7445d4698c164f962bbd85a0526ef424")},
{20, common.FromHex("8016650c7a50cf88485fd06cde52d634a89711051107f00d21fae98234f2f13d")},
{200, common.FromHex("dde92ca9812e068e6982d04b40846dc65a61a9fd4996fc0f55f2fde172a8e13c")},
{2000, common.FromHex("ab553a7f9aff82e3929c382908e30ef7dd17a332933e92ba3fe873fc661ef382")},
} {
prng := rand.New(rand.NewSource(int64(i)))
// This spongeDb is used to check the sequence of disk-db-writes
s := &spongeDb{sponge: sha3.NewLegacyKeccak256()}
db := NewDatabase(rawdb.NewDatabase(s), nil)
db := newTestDatabase(rawdb.NewDatabase(s), rawdb.HashScheme)
trie := NewEmpty(db)
// Fill the trie with elements
for i := 0; i < tc.count; i++ {
@@ -917,9 +936,9 @@ func TestCommitSequenceRandomBlobs(t *testing.T) {
}
// Flush trie -> database
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
// Flush memdb -> disk (sponge)
db.Commit(root, false)
db.Commit(root)
if got, exp := s.sponge.Sum(nil), tc.expWriteSeqHash; !bytes.Equal(got, exp) {
t.Fatalf("test %d, disk write sequence wrong:\ngot %x exp %x\n", i, got, exp)
}
@@ -930,17 +949,26 @@ func TestCommitSequenceStackTrie(t *testing.T) {
for count := 1; count < 200; count++ {
prng := rand.New(rand.NewSource(int64(count)))
// This spongeDb is used to check the sequence of disk-db-writes
s := &spongeDb{sponge: sha3.NewLegacyKeccak256(), id: "a"}
db := NewDatabase(rawdb.NewDatabase(s), nil)
s := &spongeDb{
sponge: sha3.NewLegacyKeccak256(),
id: "a",
values: make(map[string]string),
}
db := newTestDatabase(rawdb.NewDatabase(s), rawdb.HashScheme)
trie := NewEmpty(db)
// Another sponge is used for the stacktrie commits
stackTrieSponge := &spongeDb{sponge: sha3.NewLegacyKeccak256(), id: "b"}
// Another sponge is used for the stacktrie commits
stackTrieSponge := &spongeDb{
sponge: sha3.NewLegacyKeccak256(),
id: "b",
values: make(map[string]string),
}
options := NewStackTrieOptions()
options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
rawdb.WriteTrieNode(stackTrieSponge, common.Hash{}, path, hash, blob, db.Scheme())
})
stTrie := NewStackTrie(options)
// Fill the trie with elements
for i := 0; i < count; i++ {
// For the stack trie, we need to do inserts in proper order
@@ -960,13 +988,16 @@ func TestCommitSequenceStackTrie(t *testing.T) {
// Flush trie -> database
root, nodes, _ := trie.Commit(false)
// Flush memdb -> disk (sponge)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Commit(root, false)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
db.Commit(root)
s.Flush()
// And flush stacktrie -> disk
stRoot := stTrie.Commit()
if stRoot != root {
t.Fatalf("root wrong, got %x exp %x", stRoot, root)
}
stackTrieSponge.Flush()
if got, exp := stackTrieSponge.sponge.Sum(nil), s.sponge.Sum(nil); !bytes.Equal(got, exp) {
// Show the journal
t.Logf("Expected:")
@@ -989,34 +1020,47 @@ func TestCommitSequenceStackTrie(t *testing.T) {
// that even a small trie which contains a leaf will have an extension making it
// not fit into 32 bytes, rlp-encoded. However, it's still the correct thing to do.
func TestCommitSequenceSmallRoot(t *testing.T) {
s := &spongeDb{sponge: sha3.NewLegacyKeccak256(), id: "a"}
db := NewDatabase(rawdb.NewDatabase(s), nil)
s := &spongeDb{
sponge: sha3.NewLegacyKeccak256(),
id: "a",
values: make(map[string]string),
}
db := newTestDatabase(rawdb.NewDatabase(s), rawdb.HashScheme)
trie := NewEmpty(db)
// Another sponge is used for the stacktrie commits
stackTrieSponge := &spongeDb{sponge: sha3.NewLegacyKeccak256(), id: "b"}
// Another sponge is used for the stacktrie commits
stackTrieSponge := &spongeDb{
sponge: sha3.NewLegacyKeccak256(),
id: "b",
values: make(map[string]string),
}
options := NewStackTrieOptions()
options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
rawdb.WriteTrieNode(stackTrieSponge, common.Hash{}, path, hash, blob, db.Scheme())
})
stTrie := NewStackTrie(options)
// Add a single small-element to the trie(s)
key := make([]byte, 5)
key[0] = 1
trie.Update(key, []byte{0x1})
stTrie.Update(key, []byte{0x1})
// Flush trie -> database
root, nodes, _ := trie.Commit(false)
// Flush memdb -> disk (sponge)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
db.Commit(root, false)
db.Update(root, types.EmptyRootHash, trienode.NewWithNodeSet(nodes))
db.Commit(root)
// And flush stacktrie -> disk
stRoot := stTrie.Commit()
if stRoot != root {
t.Fatalf("root wrong, got %x exp %x", stRoot, root)
}
t.Logf("root: %x\n", stRoot)
s.Flush()
stackTrieSponge.Flush()
if got, exp := stackTrieSponge.sponge.Sum(nil), s.sponge.Sum(nil); !bytes.Equal(got, exp) {
t.Fatalf("test, disk write sequence wrong:\ngot %x exp %x\n", got, exp)
}
@@ -1067,7 +1111,7 @@ func BenchmarkHashFixedSize(b *testing.B) {
func benchmarkHashFixedSize(b *testing.B, addresses [][20]byte, accounts [][]byte) {
b.ReportAllocs()
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
for i := 0; i < len(addresses); i++ {
trie.MustUpdate(crypto.Keccak256(addresses[i][:]), accounts[i])
}
@@ -1118,7 +1162,7 @@ func BenchmarkCommitAfterHashFixedSize(b *testing.B) {
func benchmarkCommitAfterHashFixedSize(b *testing.B, addresses [][20]byte, accounts [][]byte) {
b.ReportAllocs()
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
trie := NewEmpty(newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.HashScheme))
for i := 0; i < len(addresses); i++ {
trie.MustUpdate(crypto.Keccak256(addresses[i][:]), accounts[i])
}
@@ -1129,60 +1173,6 @@ func benchmarkCommitAfterHashFixedSize(b *testing.B, addresses [][20]byte, accou
b.StopTimer()
}
func BenchmarkDerefRootFixedSize(b *testing.B) {
b.Run("10", func(b *testing.B) {
b.StopTimer()
acc, add := makeAccounts(20)
for i := 0; i < b.N; i++ {
benchmarkDerefRootFixedSize(b, acc, add)
}
})
b.Run("100", func(b *testing.B) {
b.StopTimer()
acc, add := makeAccounts(100)
for i := 0; i < b.N; i++ {
benchmarkDerefRootFixedSize(b, acc, add)
}
})
b.Run("1K", func(b *testing.B) {
b.StopTimer()
acc, add := makeAccounts(1000)
for i := 0; i < b.N; i++ {
benchmarkDerefRootFixedSize(b, acc, add)
}
})
b.Run("10K", func(b *testing.B) {
b.StopTimer()
acc, add := makeAccounts(10000)
for i := 0; i < b.N; i++ {
benchmarkDerefRootFixedSize(b, acc, add)
}
})
b.Run("100K", func(b *testing.B) {
b.StopTimer()
acc, add := makeAccounts(100000)
for i := 0; i < b.N; i++ {
benchmarkDerefRootFixedSize(b, acc, add)
}
})
}
func benchmarkDerefRootFixedSize(b *testing.B, addresses [][20]byte, accounts [][]byte) {
b.ReportAllocs()
triedb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
trie := NewEmpty(triedb)
for i := 0; i < len(addresses); i++ {
trie.MustUpdate(crypto.Keccak256(addresses[i][:]), accounts[i])
}
h := trie.Hash()
root, nodes, _ := trie.Commit(false)
triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
b.StartTimer()
triedb.Dereference(h)
b.StopTimer()
}
func getString(trie *Trie, k string) []byte {
return trie.MustGet([]byte(k))
}
trie/triedb/hashdb/database.go
-651
View File
@@ -1,651 +0,0 @@
// 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 hashdb
import (
"errors"
"fmt"
"reflect"
"sync"
"time"
"github.com/VictoriaMetrics/fastcache"
"github.com/ethereum/go-ethereum/common"
"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/metrics"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
var (
memcacheCleanHitMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/hit", nil)
memcacheCleanMissMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/miss", nil)
memcacheCleanReadMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/read", nil)
memcacheCleanWriteMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/write", nil)
memcacheDirtyHitMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/hit", nil)
memcacheDirtyMissMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/miss", nil)
memcacheDirtyReadMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/read", nil)
memcacheDirtyWriteMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/write", nil)
memcacheFlushTimeTimer = metrics.NewRegisteredResettingTimer("hashdb/memcache/flush/time", nil)
memcacheFlushNodesMeter = metrics.NewRegisteredMeter("hashdb/memcache/flush/nodes", nil)
memcacheFlushBytesMeter = metrics.NewRegisteredMeter("hashdb/memcache/flush/bytes", nil)
memcacheGCTimeTimer = metrics.NewRegisteredResettingTimer("hashdb/memcache/gc/time", nil)
memcacheGCNodesMeter = metrics.NewRegisteredMeter("hashdb/memcache/gc/nodes", nil)
memcacheGCBytesMeter = metrics.NewRegisteredMeter("hashdb/memcache/gc/bytes", nil)
memcacheCommitTimeTimer = metrics.NewRegisteredResettingTimer("hashdb/memcache/commit/time", nil)
memcacheCommitNodesMeter = metrics.NewRegisteredMeter("hashdb/memcache/commit/nodes", nil)
memcacheCommitBytesMeter = metrics.NewRegisteredMeter("hashdb/memcache/commit/bytes", nil)
)
// ChildResolver defines the required method to decode the provided
// trie node and iterate the children on top.
type ChildResolver interface {
ForEach(node []byte, onChild func(common.Hash))
}
// Config contains the settings for database.
type Config struct {
CleanCacheSize int // Maximum memory allowance (in bytes) for caching clean nodes
}
// Defaults is the default setting for database if it's not specified.
// Notably, clean cache is disabled explicitly,
var Defaults = &Config{
// Explicitly set clean cache size to 0 to avoid creating fastcache,
// otherwise database must be closed when it's no longer needed to
// prevent memory leak.
CleanCacheSize: 0,
}
// Database is an intermediate write layer between the trie data structures and
// the disk database. The aim is to accumulate trie writes in-memory and only
// periodically flush a couple tries to disk, garbage collecting the remainder.
type Database struct {
diskdb ethdb.Database // Persistent storage for matured trie nodes
resolver ChildResolver // The handler to resolve children of nodes
cleans *fastcache.Cache // GC friendly memory cache of clean node RLPs
dirties map[common.Hash]*cachedNode // Data and references relationships of dirty trie nodes
oldest common.Hash // Oldest tracked node, flush-list head
newest common.Hash // Newest tracked node, flush-list tail
gctime time.Duration // Time spent on garbage collection since last commit
gcnodes uint64 // Nodes garbage collected since last commit
gcsize common.StorageSize // Data storage garbage collected since last commit
flushtime time.Duration // Time spent on data flushing since last commit
flushnodes uint64 // Nodes flushed since last commit
flushsize common.StorageSize // Data storage flushed since last commit
dirtiesSize common.StorageSize // Storage size of the dirty node cache (exc. metadata)
childrenSize common.StorageSize // Storage size of the external children tracking
lock sync.RWMutex
}
// cachedNode is all the information we know about a single cached trie node
// in the memory database write layer.
type cachedNode struct {
node []byte // Encoded node blob, immutable
parents uint32 // Number of live nodes referencing this one
external map[common.Hash]struct{} // The set of external children
flushPrev common.Hash // Previous node in the flush-list
flushNext common.Hash // Next node in the flush-list
}
// cachedNodeSize is the raw size of a cachedNode data structure without any
// node data included. It's an approximate size, but should be a lot better
// than not counting them.
var cachedNodeSize = int(reflect.TypeOf(cachedNode{}).Size())
// forChildren invokes the callback for all the tracked children of this node,
// both the implicit ones from inside the node as well as the explicit ones
// from outside the node.
func (n *cachedNode) forChildren(resolver ChildResolver, onChild func(hash common.Hash)) {
for child := range n.external {
onChild(child)
}
resolver.ForEach(n.node, onChild)
}
// New initializes the hash-based node database.
func New(diskdb ethdb.Database, config *Config, resolver ChildResolver) *Database {
if config == nil {
config = Defaults
}
var cleans *fastcache.Cache
if config.CleanCacheSize > 0 {
cleans = fastcache.New(config.CleanCacheSize)
}
return &Database{
diskdb: diskdb,
resolver: resolver,
cleans: cleans,
dirties: make(map[common.Hash]*cachedNode),
}
}
// insert inserts a trie node into the memory database. All nodes inserted by
// this function will be reference tracked. This function assumes the lock is
// already held.
func (db *Database) insert(hash common.Hash, node []byte) {
// If the node's already cached, skip
if _, ok := db.dirties[hash]; ok {
return
}
memcacheDirtyWriteMeter.Mark(int64(len(node)))
// Create the cached entry for this node
entry := &cachedNode{
node: node,
flushPrev: db.newest,
}
entry.forChildren(db.resolver, func(child common.Hash) {
if c := db.dirties[child]; c != nil {
c.parents++
}
})
db.dirties[hash] = entry
// Update the flush-list endpoints
if db.oldest == (common.Hash{}) {
db.oldest, db.newest = hash, hash
} else {
db.dirties[db.newest].flushNext, db.newest = hash, hash
}
db.dirtiesSize += common.StorageSize(common.HashLength + len(node))
}
// node retrieves an encoded cached trie node from memory. If it cannot be found
// cached, the method queries the persistent database for the content.
func (db *Database) node(hash common.Hash) ([]byte, error) {
// It doesn't make sense to retrieve the metaroot
if hash == (common.Hash{}) {
return nil, errors.New("not found")
}
// Retrieve the node from the clean cache if available
if db.cleans != nil {
if enc := db.cleans.Get(nil, hash[:]); enc != nil {
memcacheCleanHitMeter.Mark(1)
memcacheCleanReadMeter.Mark(int64(len(enc)))
return enc, nil
}
}
// Retrieve the node from the dirty cache if available.
db.lock.RLock()
dirty := db.dirties[hash]
db.lock.RUnlock()
// Return the cached node if it's found in the dirty set.
// The dirty.node field is immutable and safe to read it
// even without lock guard.
if dirty != nil {
memcacheDirtyHitMeter.Mark(1)
memcacheDirtyReadMeter.Mark(int64(len(dirty.node)))
return dirty.node, nil
}
memcacheDirtyMissMeter.Mark(1)
// Content unavailable in memory, attempt to retrieve from disk
enc := rawdb.ReadLegacyTrieNode(db.diskdb, hash)
if len(enc) != 0 {
if db.cleans != nil {
db.cleans.Set(hash[:], enc)
memcacheCleanMissMeter.Mark(1)
memcacheCleanWriteMeter.Mark(int64(len(enc)))
}
return enc, nil
}
return nil, errors.New("not found")
}
// Reference adds a new reference from a parent node to a child node.
// This function is used to add reference between internal trie node
// and external node(e.g. storage trie root), all internal trie nodes
// are referenced together by database itself.
func (db *Database) Reference(child common.Hash, parent common.Hash) {
db.lock.Lock()
defer db.lock.Unlock()
db.reference(child, parent)
}
// reference is the private locked version of Reference.
func (db *Database) reference(child common.Hash, parent common.Hash) {
// If the node does not exist, it's a node pulled from disk, skip
node, ok := db.dirties[child]
if !ok {
return
}
// The reference is for state root, increase the reference counter.
if parent == (common.Hash{}) {
node.parents += 1
return
}
// The reference is for external storage trie, don't duplicate if
// the reference is already existent.
if db.dirties[parent].external == nil {
db.dirties[parent].external = make(map[common.Hash]struct{})
}
if _, ok := db.dirties[parent].external[child]; ok {
return
}
node.parents++
db.dirties[parent].external[child] = struct{}{}
db.childrenSize += common.HashLength
}
// Dereference removes an existing reference from a root node.
func (db *Database) Dereference(root common.Hash) {
// Sanity check to ensure that the meta-root is not removed
if root == (common.Hash{}) {
log.Error("Attempted to dereference the trie cache meta root")
return
}
db.lock.Lock()
defer db.lock.Unlock()
nodes, storage, start := len(db.dirties), db.dirtiesSize, time.Now()
db.dereference(root)
db.gcnodes += uint64(nodes - len(db.dirties))
db.gcsize += storage - db.dirtiesSize
db.gctime += time.Since(start)
memcacheGCTimeTimer.Update(time.Since(start))
memcacheGCBytesMeter.Mark(int64(storage - db.dirtiesSize))
memcacheGCNodesMeter.Mark(int64(nodes - len(db.dirties)))
log.Debug("Dereferenced trie from memory database", "nodes", nodes-len(db.dirties), "size", storage-db.dirtiesSize, "time", time.Since(start),
"gcnodes", db.gcnodes, "gcsize", db.gcsize, "gctime", db.gctime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
}
// dereference is the private locked version of Dereference.
func (db *Database) dereference(hash common.Hash) {
// If the node does not exist, it's a previously committed node.
node, ok := db.dirties[hash]
if !ok {
return
}
// If there are no more references to the node, delete it and cascade
if node.parents > 0 {
// This is a special cornercase where a node loaded from disk (i.e. not in the
// memcache any more) gets reinjected as a new node (short node split into full,
// then reverted into short), causing a cached node to have no parents. That is
// no problem in itself, but don't make maxint parents out of it.
node.parents--
}
if node.parents == 0 {
// Remove the node from the flush-list
switch hash {
case db.oldest:
db.oldest = node.flushNext
if node.flushNext != (common.Hash{}) {
db.dirties[node.flushNext].flushPrev = common.Hash{}
}
case db.newest:
db.newest = node.flushPrev
if node.flushPrev != (common.Hash{}) {
db.dirties[node.flushPrev].flushNext = common.Hash{}
}
default:
db.dirties[node.flushPrev].flushNext = node.flushNext
db.dirties[node.flushNext].flushPrev = node.flushPrev
}
// Dereference all children and delete the node
node.forChildren(db.resolver, func(child common.Hash) {
db.dereference(child)
})
delete(db.dirties, hash)
db.dirtiesSize -= common.StorageSize(common.HashLength + len(node.node))
if node.external != nil {
db.childrenSize -= common.StorageSize(len(node.external) * common.HashLength)
}
}
}
// Cap iteratively flushes old but still referenced trie nodes until the total
// memory usage goes below the given threshold.
func (db *Database) Cap(limit common.StorageSize) error {
db.lock.Lock()
defer db.lock.Unlock()
// Create a database batch to flush persistent data out. It is important that
// outside code doesn't see an inconsistent state (referenced data removed from
// memory cache during commit but not yet in persistent storage). This is ensured
// by only uncaching existing data when the database write finalizes.
batch := db.diskdb.NewBatch()
nodes, storage, start := len(db.dirties), db.dirtiesSize, time.Now()
// db.dirtiesSize only contains the useful data in the cache, but when reporting
// the total memory consumption, the maintenance metadata is also needed to be
// counted.
size := db.dirtiesSize + common.StorageSize(len(db.dirties)*cachedNodeSize)
size += db.childrenSize
// Keep committing nodes from the flush-list until we're below allowance
oldest := db.oldest
for size > limit && oldest != (common.Hash{}) {
// Fetch the oldest referenced node and push into the batch
node := db.dirties[oldest]
rawdb.WriteLegacyTrieNode(batch, oldest, node.node)
// If we exceeded the ideal batch size, commit and reset
if batch.ValueSize() >= ethdb.IdealBatchSize {
if err := batch.Write(); err != nil {
log.Error("Failed to write flush list to disk", "err", err)
return err
}
batch.Reset()
}
// Iterate to the next flush item, or abort if the size cap was achieved. Size
// is the total size, including the useful cached data (hash -> blob), the
// cache item metadata, as well as external children mappings.
size -= common.StorageSize(common.HashLength + len(node.node) + cachedNodeSize)
if node.external != nil {
size -= common.StorageSize(len(node.external) * common.HashLength)
}
oldest = node.flushNext
}
// Flush out any remainder data from the last batch
if err := batch.Write(); err != nil {
log.Error("Failed to write flush list to disk", "err", err)
return err
}
// Write successful, clear out the flushed data
for db.oldest != oldest {
node := db.dirties[db.oldest]
delete(db.dirties, db.oldest)
db.oldest = node.flushNext
db.dirtiesSize -= common.StorageSize(common.HashLength + len(node.node))
if node.external != nil {
db.childrenSize -= common.StorageSize(len(node.external) * common.HashLength)
}
}
if db.oldest != (common.Hash{}) {
db.dirties[db.oldest].flushPrev = common.Hash{}
}
db.flushnodes += uint64(nodes - len(db.dirties))
db.flushsize += storage - db.dirtiesSize
db.flushtime += time.Since(start)
memcacheFlushTimeTimer.Update(time.Since(start))
memcacheFlushBytesMeter.Mark(int64(storage - db.dirtiesSize))
memcacheFlushNodesMeter.Mark(int64(nodes - len(db.dirties)))
log.Debug("Persisted nodes from memory database", "nodes", nodes-len(db.dirties), "size", storage-db.dirtiesSize, "time", time.Since(start),
"flushnodes", db.flushnodes, "flushsize", db.flushsize, "flushtime", db.flushtime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
return nil
}
// Commit iterates over all the children of a particular node, writes them out
// to disk, forcefully tearing down all references in both directions. As a side
// effect, all pre-images accumulated up to this point are also written.
func (db *Database) Commit(node common.Hash, report bool) error {
db.lock.Lock()
defer db.lock.Unlock()
// Create a database batch to flush persistent data out. It is important that
// outside code doesn't see an inconsistent state (referenced data removed from
// memory cache during commit but not yet in persistent storage). This is ensured
// by only uncaching existing data when the database write finalizes.
start := time.Now()
batch := db.diskdb.NewBatch()
// Move the trie itself into the batch, flushing if enough data is accumulated
nodes, storage := len(db.dirties), db.dirtiesSize
uncacher := &cleaner{db}
if err := db.commit(node, batch, uncacher); err != nil {
log.Error("Failed to commit trie from trie database", "err", err)
return err
}
// Trie mostly committed to disk, flush any batch leftovers
if err := batch.Write(); err != nil {
log.Error("Failed to write trie to disk", "err", err)
return err
}
// Uncache any leftovers in the last batch
if err := batch.Replay(uncacher); err != nil {
return err
}
batch.Reset()
// Reset the storage counters and bumped metrics
memcacheCommitTimeTimer.Update(time.Since(start))
memcacheCommitBytesMeter.Mark(int64(storage - db.dirtiesSize))
memcacheCommitNodesMeter.Mark(int64(nodes - len(db.dirties)))
logger := log.Info
if !report {
logger = log.Debug
}
logger("Persisted trie from memory database", "nodes", nodes-len(db.dirties)+int(db.flushnodes), "size", storage-db.dirtiesSize+db.flushsize, "time", time.Since(start)+db.flushtime,
"gcnodes", db.gcnodes, "gcsize", db.gcsize, "gctime", db.gctime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
// Reset the garbage collection statistics
db.gcnodes, db.gcsize, db.gctime = 0, 0, 0
db.flushnodes, db.flushsize, db.flushtime = 0, 0, 0
return nil
}
// commit is the private locked version of Commit.
func (db *Database) commit(hash common.Hash, batch ethdb.Batch, uncacher *cleaner) error {
// If the node does not exist, it's a previously committed node
node, ok := db.dirties[hash]
if !ok {
return nil
}
var err error
// Dereference all children and delete the node
node.forChildren(db.resolver, func(child common.Hash) {
if err == nil {
err = db.commit(child, batch, uncacher)
}
})
if err != nil {
return err
}
// If we've reached an optimal batch size, commit and start over
rawdb.WriteLegacyTrieNode(batch, hash, node.node)
if batch.ValueSize() >= ethdb.IdealBatchSize {
if err := batch.Write(); err != nil {
return err
}
err := batch.Replay(uncacher)
if err != nil {
return err
}
batch.Reset()
}
return nil
}
// cleaner is a database batch replayer that takes a batch of write operations
// and cleans up the trie database from anything written to disk.
type cleaner struct {
db *Database
}
// Put reacts to database writes and implements dirty data uncaching. This is the
// post-processing step of a commit operation where the already persisted trie is
// removed from the dirty cache and moved into the clean cache. The reason behind
// the two-phase commit is to ensure data availability while moving from memory
// to disk.
func (c *cleaner) Put(key []byte, rlp []byte) error {
hash := common.BytesToHash(key)
// If the node does not exist, we're done on this path
node, ok := c.db.dirties[hash]
if !ok {
return nil
}
// Node still exists, remove it from the flush-list
switch hash {
case c.db.oldest:
c.db.oldest = node.flushNext
if node.flushNext != (common.Hash{}) {
c.db.dirties[node.flushNext].flushPrev = common.Hash{}
}
case c.db.newest:
c.db.newest = node.flushPrev
if node.flushPrev != (common.Hash{}) {
c.db.dirties[node.flushPrev].flushNext = common.Hash{}
}
default:
c.db.dirties[node.flushPrev].flushNext = node.flushNext
c.db.dirties[node.flushNext].flushPrev = node.flushPrev
}
// Remove the node from the dirty cache
delete(c.db.dirties, hash)
c.db.dirtiesSize -= common.StorageSize(common.HashLength + len(node.node))
if node.external != nil {
c.db.childrenSize -= common.StorageSize(len(node.external) * common.HashLength)
}
// Move the flushed node into the clean cache to prevent insta-reloads
if c.db.cleans != nil {
c.db.cleans.Set(hash[:], rlp)
memcacheCleanWriteMeter.Mark(int64(len(rlp)))
}
return nil
}
func (c *cleaner) Delete(key []byte) error {
panic("not implemented")
}
// Initialized returns an indicator if state data is already initialized
// in hash-based scheme by checking the presence of genesis state.
func (db *Database) Initialized(genesisRoot common.Hash) bool {
return rawdb.HasLegacyTrieNode(db.diskdb, genesisRoot)
}
// Update inserts the dirty nodes in provided nodeset into database and link the
// account trie with multiple storage tries if necessary.
func (db *Database) Update(root common.Hash, parent common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
// Ensure the parent state is present and signal a warning if not.
if parent != types.EmptyRootHash {
if blob, _ := db.node(parent); len(blob) == 0 {
log.Error("parent state is not present")
}
}
db.lock.Lock()
defer db.lock.Unlock()
// Insert dirty nodes into the database. In the same tree, it must be
// ensured that children are inserted first, then parent so that children
// can be linked with their parent correctly.
//
// Note, the storage tries must be flushed before the account trie to
// retain the invariant that children go into the dirty cache first.
var order []common.Hash
for owner := range nodes.Sets {
if owner == (common.Hash{}) {
continue
}
order = append(order, owner)
}
if _, ok := nodes.Sets[common.Hash{}]; ok {
order = append(order, common.Hash{})
}
for _, owner := range order {
subset := nodes.Sets[owner]
subset.ForEachWithOrder(func(path string, n *trienode.Node) {
if n.IsDeleted() {
return // ignore deletion
}
db.insert(n.Hash, n.Blob)
})
}
// Link up the account trie and storage trie if the node points
// to an account trie leaf.
if set, present := nodes.Sets[common.Hash{}]; present {
for _, n := range set.Leaves {
var account types.StateAccount
if err := rlp.DecodeBytes(n.Blob, &account); err != nil {
return err
}
if account.Root != types.EmptyRootHash {
db.reference(account.Root, n.Parent)
}
}
}
return nil
}
// Size returns the current storage size of the memory cache in front of the
// persistent database layer.
//
// The first return will always be 0, representing the memory stored in unbounded
// diff layers above the dirty cache. This is only available in pathdb.
func (db *Database) Size() (common.StorageSize, common.StorageSize) {
db.lock.RLock()
defer db.lock.RUnlock()
// db.dirtiesSize only contains the useful data in the cache, but when reporting
// the total memory consumption, the maintenance metadata is also needed to be
// counted.
var metadataSize = common.StorageSize(len(db.dirties) * cachedNodeSize)
return 0, db.dirtiesSize + db.childrenSize + metadataSize
}
// Close closes the trie database and releases all held resources.
func (db *Database) Close() error {
if db.cleans != nil {
db.cleans.Reset()
db.cleans = nil
}
return nil
}
// Scheme returns the node scheme used in the database.
func (db *Database) Scheme() string {
return rawdb.HashScheme
}
// Reader retrieves a node reader belonging to the given state root.
// An error will be returned if the requested state is not available.
func (db *Database) Reader(root common.Hash) (*reader, error) {
if _, err := db.node(root); err != nil {
return nil, fmt.Errorf("state %#x is not available, %v", root, err)
}
return &reader{db: db}, nil
}
// reader is a state reader of Database which implements the Reader interface.
type reader struct {
db *Database
}
// Node retrieves the trie node with the given node hash. No error will be
// returned if the node is not found.
func (reader *reader) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
blob, _ := reader.db.node(hash)
return blob, nil
}
trie/triedb/pathdb/database.go
-485
View File
@@ -1,485 +0,0 @@
// Copyright 2022 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 pathdb
import (
"errors"
"fmt"
"io"
"sync"
"time"
"github.com/ethereum/go-ethereum/common"
"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/params"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
const (
// maxDiffLayers is the maximum diff layers allowed in the layer tree.
maxDiffLayers = 128
// defaultCleanSize is the default memory allowance of clean cache.
defaultCleanSize = 16 * 1024 * 1024
// maxBufferSize is the maximum memory allowance of node buffer.
// Too large nodebuffer will cause the system to pause for a long
// time when write happens. Also, the largest batch that pebble can
// support is 4GB, node will panic if batch size exceeds this limit.
maxBufferSize = 256 * 1024 * 1024
// DefaultBufferSize is the default memory allowance of node buffer
// that aggregates the writes from above until it's flushed into the
// disk. It's meant to be used once the initial sync is finished.
// Do not increase the buffer size arbitrarily, otherwise the system
// pause time will increase when the database writes happen.
DefaultBufferSize = 64 * 1024 * 1024
)
// layer is the interface implemented by all state layers which includes some
// public methods and some additional methods for internal usage.
type layer interface {
// Node retrieves the trie node with the node info. An error will be returned
// if the read operation exits abnormally. For example, if the layer is already
// stale, or the associated state is regarded as corrupted. Notably, no error
// will be returned if the requested node is not found in database.
Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error)
// rootHash returns the root hash for which this layer was made.
rootHash() common.Hash
// stateID returns the associated state id of layer.
stateID() uint64
// parentLayer returns the subsequent layer of it, or nil if the disk was reached.
parentLayer() layer
// update creates a new layer on top of the existing layer diff tree with
// the provided dirty trie nodes along with the state change set.
//
// Note, the maps are retained by the method to avoid copying everything.
update(root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer
// journal commits an entire diff hierarchy to disk into a single journal entry.
// This is meant to be used during shutdown to persist the layer without
// flattening everything down (bad for reorgs).
journal(w io.Writer) error
}
// Config contains the settings for database.
type Config struct {
StateHistory uint64 // Number of recent blocks to maintain state history for
CleanCacheSize int // Maximum memory allowance (in bytes) for caching clean nodes
DirtyCacheSize int // Maximum memory allowance (in bytes) for caching dirty nodes
ReadOnly bool // Flag whether the database is opened in read only mode.
}
// sanitize checks the provided user configurations and changes anything that's
// unreasonable or unworkable.
func (c *Config) sanitize() *Config {
conf := *c
if conf.DirtyCacheSize > maxBufferSize {
log.Warn("Sanitizing invalid node buffer size", "provided", common.StorageSize(conf.DirtyCacheSize), "updated", common.StorageSize(maxBufferSize))
conf.DirtyCacheSize = maxBufferSize
}
return &conf
}
// Defaults contains default settings for Ethereum mainnet.
var Defaults = &Config{
StateHistory: params.FullImmutabilityThreshold,
CleanCacheSize: defaultCleanSize,
DirtyCacheSize: DefaultBufferSize,
}
// ReadOnly is the config in order to open database in read only mode.
var ReadOnly = &Config{ReadOnly: true}
// Database is a multiple-layered structure for maintaining in-memory trie nodes.
// It consists of one persistent base layer backed by a key-value store, on top
// of which arbitrarily many in-memory diff layers are stacked. 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, a batch of reverse diffs can
// be applied to rollback. The deepest reorg that can be handled depends on the
// amount of state histories tracked in the disk.
//
// At most one readable and writable database can be opened at the same time in
// the whole system which ensures that only one database writer can operate disk
// state. Unexpected open operations can cause the system to panic.
type Database struct {
// readOnly is the flag whether the mutation is allowed to be applied.
// It will be set automatically when the database is journaled during
// the shutdown to reject all following unexpected mutations.
readOnly bool // Flag if database is opened in read only mode
waitSync bool // Flag if database is deactivated due to initial state sync
bufferSize int // Memory allowance (in bytes) for caching dirty nodes
config *Config // Configuration for database
diskdb ethdb.Database // Persistent storage for matured trie nodes
tree *layerTree // The group for all known layers
freezer *rawdb.ResettableFreezer // Freezer for storing trie histories, nil possible in tests
lock sync.RWMutex // Lock to prevent mutations from happening at the same time
}
// New attempts to load an already existing layer from a persistent key-value
// store (with a number of memory layers from a journal). If the journal is not
// matched with the base persistent layer, all the recorded diff layers are discarded.
func New(diskdb ethdb.Database, config *Config) *Database {
if config == nil {
config = Defaults
}
config = config.sanitize()
db := &Database{
readOnly: config.ReadOnly,
bufferSize: config.DirtyCacheSize,
config: config,
diskdb: diskdb,
}
// Construct the layer tree by resolving the in-disk singleton state
// and in-memory layer journal.
db.tree = newLayerTree(db.loadLayers())
// Open the freezer for state history if the passed database contains an
// ancient store. Otherwise, all the relevant functionalities are disabled.
//
// Because the freezer can only be opened once at the same time, this
// mechanism also ensures that at most one **non-readOnly** database
// is opened at the same time to prevent accidental mutation.
if ancient, err := diskdb.AncientDatadir(); err == nil && ancient != "" && !db.readOnly {
freezer, err := rawdb.NewStateFreezer(ancient, false)
if err != nil {
log.Crit("Failed to open state history freezer", "err", err)
}
db.freezer = freezer
diskLayerID := db.tree.bottom().stateID()
if diskLayerID == 0 {
// Reset the entire state histories in case the trie database is
// not initialized yet, as these state histories are not expected.
frozen, err := db.freezer.Ancients()
if err != nil {
log.Crit("Failed to retrieve head of state history", "err", err)
}
if frozen != 0 {
err := db.freezer.Reset()
if err != nil {
log.Crit("Failed to reset state histories", "err", err)
}
log.Info("Truncated extraneous state history")
}
} else {
// Truncate the extra state histories above in freezer in case
// it's not aligned with the disk layer.
pruned, err := truncateFromHead(db.diskdb, freezer, diskLayerID)
if err != nil {
log.Crit("Failed to truncate extra state histories", "err", err)
}
if pruned != 0 {
log.Warn("Truncated extra state histories", "number", pruned)
}
}
}
// Disable database in case node is still in the initial state sync stage.
if rawdb.ReadSnapSyncStatusFlag(diskdb) == rawdb.StateSyncRunning && !db.readOnly {
if err := db.Disable(); err != nil {
log.Crit("Failed to disable database", "err", err) // impossible to happen
}
}
log.Warn("Path-based state scheme is an experimental feature")
return db
}
// Reader retrieves a layer belonging to the given state root.
func (db *Database) Reader(root common.Hash) (layer, error) {
l := db.tree.get(root)
if l == nil {
return nil, fmt.Errorf("state %#x is not available", root)
}
return l, nil
}
// Update adds a new layer 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). Apart
// from that this function will flatten the extra diff layers at bottom into disk
// to only keep 128 diff layers in memory by default.
//
// The passed in maps(nodes, states) will be retained to avoid copying everything.
// Therefore, these maps must not be changed afterwards.
func (db *Database) Update(root common.Hash, parentRoot common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
// Hold the lock to prevent concurrent mutations.
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the mutation is not allowed.
if err := db.modifyAllowed(); err != nil {
return err
}
if err := db.tree.add(root, parentRoot, block, nodes, states); err != nil {
return err
}
// Keep 128 diff layers in the memory, persistent layer is 129th.
// - head layer is paired with HEAD state
// - head-1 layer is paired with HEAD-1 state
// - head-127 layer(bottom-most diff layer) is paired with HEAD-127 state
// - head-128 layer(disk layer) is paired with HEAD-128 state
return db.tree.cap(root, maxDiffLayers)
}
// Commit traverses downwards the layer tree from a specified layer with the
// provided state root and all the layers below are flattened downwards. It
// can be used alone and mostly for test purposes.
func (db *Database) Commit(root common.Hash, report bool) error {
// Hold the lock to prevent concurrent mutations.
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the mutation is not allowed.
if err := db.modifyAllowed(); err != nil {
return err
}
return db.tree.cap(root, 0)
}
// Disable deactivates the database and invalidates all available state layers
// as stale to prevent access to the persistent state, which is in the syncing
// stage.
func (db *Database) Disable() error {
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the database is in read only mode.
if db.readOnly {
return errDatabaseReadOnly
}
// Prevent duplicated disable operation.
if db.waitSync {
log.Error("Reject duplicated disable operation")
return nil
}
db.waitSync = true
// Mark the disk layer as stale to prevent access to persistent state.
db.tree.bottom().markStale()
// Write the initial sync flag to persist it across restarts.
rawdb.WriteSnapSyncStatusFlag(db.diskdb, rawdb.StateSyncRunning)
log.Info("Disabled trie database due to state sync")
return nil
}
// Enable activates database and resets the state tree with the provided persistent
// state root once the state sync is finished.
func (db *Database) Enable(root common.Hash) error {
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the database is in read only mode.
if db.readOnly {
return errDatabaseReadOnly
}
// Ensure the provided state root matches the stored one.
root = types.TrieRootHash(root)
_, stored := rawdb.ReadAccountTrieNode(db.diskdb, nil)
if stored != root {
return fmt.Errorf("state root mismatch: stored %x, synced %x", stored, root)
}
// Drop the stale state journal in persistent database and
// reset the persistent state id back to zero.
batch := db.diskdb.NewBatch()
rawdb.DeleteTrieJournal(batch)
rawdb.WritePersistentStateID(batch, 0)
if err := batch.Write(); err != nil {
return err
}
// Clean up all state histories in freezer. Theoretically
// all root->id mappings should be removed as well. Since
// mappings can be huge and might take a while to clear
// them, just leave them in disk and wait for overwriting.
if db.freezer != nil {
if err := db.freezer.Reset(); err != nil {
return err
}
}
// Re-construct a new disk layer backed by persistent state
// with **empty clean cache and node buffer**.
db.tree.reset(newDiskLayer(root, 0, db, nil, newNodeBuffer(db.bufferSize, nil, 0)))
// Re-enable the database as the final step.
db.waitSync = false
rawdb.WriteSnapSyncStatusFlag(db.diskdb, rawdb.StateSyncFinished)
log.Info("Rebuilt trie database", "root", root)
return nil
}
// Recover rollbacks the database to a specified historical point.
// The state is supported as the rollback destination only if it's
// canonical state and the corresponding trie histories are existent.
func (db *Database) Recover(root common.Hash, loader triestate.TrieLoader) error {
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if rollback operation is not supported.
if err := db.modifyAllowed(); err != nil {
return err
}
if db.freezer == nil {
return errors.New("state rollback is non-supported")
}
// Short circuit if the target state is not recoverable.
root = types.TrieRootHash(root)
if !db.Recoverable(root) {
return errStateUnrecoverable
}
// Apply the state histories upon the disk layer in order.
var (
start = time.Now()
dl = db.tree.bottom()
)
for dl.rootHash() != root {
h, err := readHistory(db.freezer, dl.stateID())
if err != nil {
return err
}
dl, err = dl.revert(h, loader)
if err != nil {
return err
}
// reset layer with newly created disk layer. It must be
// done after each revert operation, otherwise the new
// disk layer won't be accessible from outside.
db.tree.reset(dl)
}
rawdb.DeleteTrieJournal(db.diskdb)
_, err := truncateFromHead(db.diskdb, db.freezer, dl.stateID())
if err != nil {
return err
}
log.Debug("Recovered state", "root", root, "elapsed", common.PrettyDuration(time.Since(start)))
return nil
}
// Recoverable returns the indicator if the specified state is recoverable.
func (db *Database) Recoverable(root common.Hash) bool {
// Ensure the requested state is a known state.
root = types.TrieRootHash(root)
id := rawdb.ReadStateID(db.diskdb, root)
if id == nil {
return false
}
// Recoverable state must below the disk layer. The recoverable
// state only refers the state that is currently not available,
// but can be restored by applying state history.
dl := db.tree.bottom()
if *id >= dl.stateID() {
return false
}
// Ensure the requested state is a canonical state and all state
// histories in range [id+1, disklayer.ID] are present and complete.
parent := root
return checkHistories(db.freezer, *id+1, dl.stateID()-*id, func(m *meta) error {
if m.parent != parent {
return errors.New("unexpected state history")
}
if len(m.incomplete) > 0 {
return errors.New("incomplete state history")
}
parent = m.root
return nil
}) == nil
}
// Close closes the trie database and the held freezer.
func (db *Database) Close() error {
db.lock.Lock()
defer db.lock.Unlock()
// Set the database to read-only mode to prevent all
// following mutations.
db.readOnly = true
// Release the memory held by clean cache.
db.tree.bottom().resetCache()
// Close the attached state history freezer.
if db.freezer == nil {
return nil
}
return db.freezer.Close()
}
// Size returns the current storage size of the memory cache in front of the
// persistent database layer.
func (db *Database) Size() (diffs common.StorageSize, nodes common.StorageSize) {
db.tree.forEach(func(layer layer) {
if diff, ok := layer.(*diffLayer); ok {
diffs += common.StorageSize(diff.memory)
}
if disk, ok := layer.(*diskLayer); ok {
nodes += disk.size()
}
})
return diffs, nodes
}
// Initialized returns an indicator if the state data is already
// initialized in path-based scheme.
func (db *Database) Initialized(genesisRoot common.Hash) bool {
var inited bool
db.tree.forEach(func(layer layer) {
if layer.rootHash() != types.EmptyRootHash {
inited = true
}
})
if !inited {
inited = rawdb.ReadSnapSyncStatusFlag(db.diskdb) != rawdb.StateSyncUnknown
}
return inited
}
// SetBufferSize sets the node buffer size to the provided value(in bytes).
func (db *Database) SetBufferSize(size int) error {
db.lock.Lock()
defer db.lock.Unlock()
if size > maxBufferSize {
log.Info("Capped node buffer size", "provided", common.StorageSize(size), "adjusted", common.StorageSize(maxBufferSize))
size = maxBufferSize
}
db.bufferSize = size
return db.tree.bottom().setBufferSize(db.bufferSize)
}
// Scheme returns the node scheme used in the database.
func (db *Database) Scheme() string {
return rawdb.PathScheme
}
// modifyAllowed returns the indicator if mutation is allowed. This function
// assumes the db.lock is already held.
func (db *Database) modifyAllowed() error {
if db.readOnly {
return errDatabaseReadOnly
}
if db.waitSync {
return errDatabaseWaitSync
}
return nil
}
trie/triedb/pathdb/database_test.go
-608
View File
@@ -1,608 +0,0 @@
// Copyright 2022 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 pathdb
import (
"bytes"
"errors"
"fmt"
"math/rand"
"testing"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/testutil"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
"github.com/holiman/uint256"
)
func updateTrie(addrHash common.Hash, root common.Hash, dirties, cleans map[common.Hash][]byte) (common.Hash, *trienode.NodeSet) {
h, err := newTestHasher(addrHash, root, cleans)
if err != nil {
panic(fmt.Errorf("failed to create hasher, err: %w", err))
}
for key, val := range dirties {
if len(val) == 0 {
h.Delete(key.Bytes())
} else {
h.Update(key.Bytes(), val)
}
}
root, nodes, _ := h.Commit(false)
return root, nodes
}
func generateAccount(storageRoot common.Hash) types.StateAccount {
return types.StateAccount{
Nonce: uint64(rand.Intn(100)),
Balance: uint256.NewInt(rand.Uint64()),
CodeHash: testutil.RandBytes(32),
Root: storageRoot,
}
}
const (
createAccountOp int = iota
modifyAccountOp
deleteAccountOp
opLen
)
type genctx struct {
accounts map[common.Hash][]byte
storages map[common.Hash]map[common.Hash][]byte
accountOrigin map[common.Address][]byte
storageOrigin map[common.Address]map[common.Hash][]byte
nodes *trienode.MergedNodeSet
}
func newCtx() *genctx {
return &genctx{
accounts: make(map[common.Hash][]byte),
storages: make(map[common.Hash]map[common.Hash][]byte),
accountOrigin: make(map[common.Address][]byte),
storageOrigin: make(map[common.Address]map[common.Hash][]byte),
nodes: trienode.NewMergedNodeSet(),
}
}
type tester struct {
db *Database
roots []common.Hash
preimages map[common.Hash]common.Address
accounts map[common.Hash][]byte
storages map[common.Hash]map[common.Hash][]byte
// state snapshots
snapAccounts map[common.Hash]map[common.Hash][]byte
snapStorages map[common.Hash]map[common.Hash]map[common.Hash][]byte
}
func newTester(t *testing.T, historyLimit uint64) *tester {
var (
disk, _ = rawdb.NewDatabaseWithFreezer(rawdb.NewMemoryDatabase(), t.TempDir(), "", false)
db = New(disk, &Config{
StateHistory: historyLimit,
CleanCacheSize: 256 * 1024,
DirtyCacheSize: 256 * 1024,
})
obj = &tester{
db: db,
preimages: make(map[common.Hash]common.Address),
accounts: make(map[common.Hash][]byte),
storages: make(map[common.Hash]map[common.Hash][]byte),
snapAccounts: make(map[common.Hash]map[common.Hash][]byte),
snapStorages: make(map[common.Hash]map[common.Hash]map[common.Hash][]byte),
}
)
for i := 0; i < 2*128; i++ {
var parent = types.EmptyRootHash
if len(obj.roots) != 0 {
parent = obj.roots[len(obj.roots)-1]
}
root, nodes, states := obj.generate(parent)
if err := db.Update(root, parent, uint64(i), nodes, states); err != nil {
panic(fmt.Errorf("failed to update state changes, err: %w", err))
}
obj.roots = append(obj.roots, root)
}
return obj
}
func (t *tester) release() {
t.db.Close()
t.db.diskdb.Close()
}
func (t *tester) randAccount() (common.Address, []byte) {
for addrHash, account := range t.accounts {
return t.preimages[addrHash], account
}
return common.Address{}, nil
}
func (t *tester) generateStorage(ctx *genctx, addr common.Address) common.Hash {
var (
addrHash = crypto.Keccak256Hash(addr.Bytes())
storage = make(map[common.Hash][]byte)
origin = make(map[common.Hash][]byte)
)
for i := 0; i < 10; i++ {
v, _ := rlp.EncodeToBytes(common.TrimLeftZeroes(testutil.RandBytes(32)))
hash := testutil.RandomHash()
storage[hash] = v
origin[hash] = nil
}
root, set := updateTrie(addrHash, types.EmptyRootHash, storage, nil)
ctx.storages[addrHash] = storage
ctx.storageOrigin[addr] = origin
ctx.nodes.Merge(set)
return root
}
func (t *tester) mutateStorage(ctx *genctx, addr common.Address, root common.Hash) common.Hash {
var (
addrHash = crypto.Keccak256Hash(addr.Bytes())
storage = make(map[common.Hash][]byte)
origin = make(map[common.Hash][]byte)
)
for hash, val := range t.storages[addrHash] {
origin[hash] = val
storage[hash] = nil
if len(origin) == 3 {
break
}
}
for i := 0; i < 3; i++ {
v, _ := rlp.EncodeToBytes(common.TrimLeftZeroes(testutil.RandBytes(32)))
hash := testutil.RandomHash()
storage[hash] = v
origin[hash] = nil
}
root, set := updateTrie(crypto.Keccak256Hash(addr.Bytes()), root, storage, t.storages[addrHash])
ctx.storages[addrHash] = storage
ctx.storageOrigin[addr] = origin
ctx.nodes.Merge(set)
return root
}
func (t *tester) clearStorage(ctx *genctx, addr common.Address, root common.Hash) common.Hash {
var (
addrHash = crypto.Keccak256Hash(addr.Bytes())
storage = make(map[common.Hash][]byte)
origin = make(map[common.Hash][]byte)
)
for hash, val := range t.storages[addrHash] {
origin[hash] = val
storage[hash] = nil
}
root, set := updateTrie(addrHash, root, storage, t.storages[addrHash])
if root != types.EmptyRootHash {
panic("failed to clear storage trie")
}
ctx.storages[addrHash] = storage
ctx.storageOrigin[addr] = origin
ctx.nodes.Merge(set)
return root
}
func (t *tester) generate(parent common.Hash) (common.Hash, *trienode.MergedNodeSet, *triestate.Set) {
var (
ctx = newCtx()
dirties = make(map[common.Hash]struct{})
)
for i := 0; i < 20; i++ {
switch rand.Intn(opLen) {
case createAccountOp:
// account creation
addr := testutil.RandomAddress()
addrHash := crypto.Keccak256Hash(addr.Bytes())
if _, ok := t.accounts[addrHash]; ok {
continue
}
if _, ok := dirties[addrHash]; ok {
continue
}
dirties[addrHash] = struct{}{}
root := t.generateStorage(ctx, addr)
ctx.accounts[addrHash] = types.SlimAccountRLP(generateAccount(root))
ctx.accountOrigin[addr] = nil
t.preimages[addrHash] = addr
case modifyAccountOp:
// account mutation
addr, account := t.randAccount()
if addr == (common.Address{}) {
continue
}
addrHash := crypto.Keccak256Hash(addr.Bytes())
if _, ok := dirties[addrHash]; ok {
continue
}
dirties[addrHash] = struct{}{}
acct, _ := types.FullAccount(account)
stRoot := t.mutateStorage(ctx, addr, acct.Root)
newAccount := types.SlimAccountRLP(generateAccount(stRoot))
ctx.accounts[addrHash] = newAccount
ctx.accountOrigin[addr] = account
case deleteAccountOp:
// account deletion
addr, account := t.randAccount()
if addr == (common.Address{}) {
continue
}
addrHash := crypto.Keccak256Hash(addr.Bytes())
if _, ok := dirties[addrHash]; ok {
continue
}
dirties[addrHash] = struct{}{}
acct, _ := types.FullAccount(account)
if acct.Root != types.EmptyRootHash {
t.clearStorage(ctx, addr, acct.Root)
}
ctx.accounts[addrHash] = nil
ctx.accountOrigin[addr] = account
}
}
root, set := updateTrie(common.Hash{}, parent, ctx.accounts, t.accounts)
ctx.nodes.Merge(set)
// Save state snapshot before commit
t.snapAccounts[parent] = copyAccounts(t.accounts)
t.snapStorages[parent] = copyStorages(t.storages)
// Commit all changes to live state set
for addrHash, account := range ctx.accounts {
if len(account) == 0 {
delete(t.accounts, addrHash)
} else {
t.accounts[addrHash] = account
}
}
for addrHash, slots := range ctx.storages {
if _, ok := t.storages[addrHash]; !ok {
t.storages[addrHash] = make(map[common.Hash][]byte)
}
for sHash, slot := range slots {
if len(slot) == 0 {
delete(t.storages[addrHash], sHash)
} else {
t.storages[addrHash][sHash] = slot
}
}
}
return root, ctx.nodes, triestate.New(ctx.accountOrigin, ctx.storageOrigin, nil)
}
// lastRoot returns the latest root hash, or empty if nothing is cached.
func (t *tester) lastHash() common.Hash {
if len(t.roots) == 0 {
return common.Hash{}
}
return t.roots[len(t.roots)-1]
}
func (t *tester) verifyState(root common.Hash) error {
reader, err := t.db.Reader(root)
if err != nil {
return err
}
_, err = reader.Node(common.Hash{}, nil, root)
if err != nil {
return errors.New("root node is not available")
}
for addrHash, account := range t.snapAccounts[root] {
blob, err := reader.Node(common.Hash{}, addrHash.Bytes(), crypto.Keccak256Hash(account))
if err != nil || !bytes.Equal(blob, account) {
return fmt.Errorf("account is mismatched: %w", err)
}
}
for addrHash, slots := range t.snapStorages[root] {
for hash, slot := range slots {
blob, err := reader.Node(addrHash, hash.Bytes(), crypto.Keccak256Hash(slot))
if err != nil || !bytes.Equal(blob, slot) {
return fmt.Errorf("slot is mismatched: %w", err)
}
}
}
return nil
}
func (t *tester) verifyHistory() error {
bottom := t.bottomIndex()
for i, root := range t.roots {
// The state history related to the state above disk layer should not exist.
if i > bottom {
_, err := readHistory(t.db.freezer, uint64(i+1))
if err == nil {
return errors.New("unexpected state history")
}
continue
}
// The state history related to the state below or equal to the disk layer
// should exist.
obj, err := readHistory(t.db.freezer, uint64(i+1))
if err != nil {
return err
}
parent := types.EmptyRootHash
if i != 0 {
parent = t.roots[i-1]
}
if obj.meta.parent != parent {
return fmt.Errorf("unexpected parent, want: %x, got: %x", parent, obj.meta.parent)
}
if obj.meta.root != root {
return fmt.Errorf("unexpected root, want: %x, got: %x", root, obj.meta.root)
}
}
return nil
}
// bottomIndex returns the index of current disk layer.
func (t *tester) bottomIndex() int {
bottom := t.db.tree.bottom()
for i := 0; i < len(t.roots); i++ {
if t.roots[i] == bottom.rootHash() {
return i
}
}
return -1
}
func TestDatabaseRollback(t *testing.T) {
// Verify state histories
tester := newTester(t, 0)
defer tester.release()
if err := tester.verifyHistory(); err != nil {
t.Fatalf("Invalid state history, err: %v", err)
}
// Revert database from top to bottom
for i := tester.bottomIndex(); i >= 0; i-- {
root := tester.roots[i]
parent := types.EmptyRootHash
if i > 0 {
parent = tester.roots[i-1]
}
loader := newHashLoader(tester.snapAccounts[root], tester.snapStorages[root])
if err := tester.db.Recover(parent, loader); err != nil {
t.Fatalf("Failed to revert db, err: %v", err)
}
tester.verifyState(parent)
}
if tester.db.tree.len() != 1 {
t.Fatal("Only disk layer is expected")
}
}
func TestDatabaseRecoverable(t *testing.T) {
var (
tester = newTester(t, 0)
index = tester.bottomIndex()
)
defer tester.release()
var cases = []struct {
root common.Hash
expect bool
}{
// Unknown state should be unrecoverable
{common.Hash{0x1}, false},
// Initial state should be recoverable
{types.EmptyRootHash, true},
// Initial state should be recoverable
{common.Hash{}, true},
// Layers below current disk layer are recoverable
{tester.roots[index-1], true},
// Disklayer itself is not recoverable, since it's
// available for accessing.
{tester.roots[index], false},
// Layers above current disk layer are not recoverable
// since they are available for accessing.
{tester.roots[index+1], false},
}
for i, c := range cases {
result := tester.db.Recoverable(c.root)
if result != c.expect {
t.Fatalf("case: %d, unexpected result, want %t, got %t", i, c.expect, result)
}
}
}
func TestDisable(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
_, stored := rawdb.ReadAccountTrieNode(tester.db.diskdb, nil)
if err := tester.db.Disable(); err != nil {
t.Fatal("Failed to deactivate database")
}
if err := tester.db.Enable(types.EmptyRootHash); err == nil {
t.Fatalf("Invalid activation should be rejected")
}
if err := tester.db.Enable(stored); err != nil {
t.Fatal("Failed to activate database")
}
// Ensure journal is deleted from disk
if blob := rawdb.ReadTrieJournal(tester.db.diskdb); len(blob) != 0 {
t.Fatal("Failed to clean journal")
}
// Ensure all trie histories are removed
n, err := tester.db.freezer.Ancients()
if err != nil {
t.Fatal("Failed to clean state history")
}
if n != 0 {
t.Fatal("Failed to clean state history")
}
// Verify layer tree structure, single disk layer is expected
if tester.db.tree.len() != 1 {
t.Fatalf("Extra layer kept %d", tester.db.tree.len())
}
if tester.db.tree.bottom().rootHash() != stored {
t.Fatalf("Root hash is not matched exp %x got %x", stored, tester.db.tree.bottom().rootHash())
}
}
func TestCommit(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
if err := tester.db.Commit(tester.lastHash(), false); err != nil {
t.Fatalf("Failed to cap database, err: %v", err)
}
// Verify layer tree structure, single disk layer is expected
if tester.db.tree.len() != 1 {
t.Fatal("Layer tree structure is invalid")
}
if tester.db.tree.bottom().rootHash() != tester.lastHash() {
t.Fatal("Layer tree structure is invalid")
}
// Verify states
if err := tester.verifyState(tester.lastHash()); err != nil {
t.Fatalf("State is invalid, err: %v", err)
}
// Verify state histories
if err := tester.verifyHistory(); err != nil {
t.Fatalf("State history is invalid, err: %v", err)
}
}
func TestJournal(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
if err := tester.db.Journal(tester.lastHash()); err != nil {
t.Errorf("Failed to journal, err: %v", err)
}
tester.db.Close()
tester.db = New(tester.db.diskdb, nil)
// Verify states including disk layer and all diff on top.
for i := 0; i < len(tester.roots); i++ {
if i >= tester.bottomIndex() {
if err := tester.verifyState(tester.roots[i]); err != nil {
t.Fatalf("Invalid state, err: %v", err)
}
continue
}
if err := tester.verifyState(tester.roots[i]); err == nil {
t.Fatal("Unexpected state")
}
}
}
func TestCorruptedJournal(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
if err := tester.db.Journal(tester.lastHash()); err != nil {
t.Errorf("Failed to journal, err: %v", err)
}
tester.db.Close()
_, root := rawdb.ReadAccountTrieNode(tester.db.diskdb, nil)
// Mutate the journal in disk, it should be regarded as invalid
blob := rawdb.ReadTrieJournal(tester.db.diskdb)
blob[0] = 1
rawdb.WriteTrieJournal(tester.db.diskdb, blob)
// Verify states, all not-yet-written states should be discarded
tester.db = New(tester.db.diskdb, nil)
for i := 0; i < len(tester.roots); i++ {
if tester.roots[i] == root {
if err := tester.verifyState(root); err != nil {
t.Fatalf("Disk state is corrupted, err: %v", err)
}
continue
}
if err := tester.verifyState(tester.roots[i]); err == nil {
t.Fatal("Unexpected state")
}
}
}
// TestTailTruncateHistory function is designed to test a specific edge case where,
// when history objects are removed from the end, it should trigger a state flush
// if the ID of the new tail object is even higher than the persisted state ID.
//
// For example, let's say the ID of the persistent state is 10, and the current
// history objects range from ID(5) to ID(15). As we accumulate six more objects,
// the history will expand to cover ID(11) to ID(21). ID(11) then becomes the
// oldest history object, and its ID is even higher than the stored state.
//
// In this scenario, it is mandatory to update the persistent state before
// truncating the tail histories. This ensures that the ID of the persistent state
// always falls within the range of [oldest-history-id, latest-history-id].
func TestTailTruncateHistory(t *testing.T) {
tester := newTester(t, 10)
defer tester.release()
tester.db.Close()
tester.db = New(tester.db.diskdb, &Config{StateHistory: 10})
head, err := tester.db.freezer.Ancients()
if err != nil {
t.Fatalf("Failed to obtain freezer head")
}
stored := rawdb.ReadPersistentStateID(tester.db.diskdb)
if head != stored {
t.Fatalf("Failed to truncate excess history object above, stored: %d, head: %d", stored, head)
}
}
// copyAccounts returns a deep-copied account set of the provided one.
func copyAccounts(set map[common.Hash][]byte) map[common.Hash][]byte {
copied := make(map[common.Hash][]byte, len(set))
for key, val := range set {
copied[key] = common.CopyBytes(val)
}
return copied
}
// copyStorages returns a deep-copied storage set of the provided one.
func copyStorages(set map[common.Hash]map[common.Hash][]byte) map[common.Hash]map[common.Hash][]byte {
copied := make(map[common.Hash]map[common.Hash][]byte, len(set))
for addrHash, subset := range set {
copied[addrHash] = make(map[common.Hash][]byte, len(subset))
for key, val := range subset {
copied[addrHash][key] = common.CopyBytes(val)
}
}
return copied
}
trie/triedb/pathdb/difflayer.go
-174
View File
@@ -1,174 +0,0 @@
// Copyright 2022 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 pathdb
import (
"fmt"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// diffLayer represents a collection of modifications made to the in-memory tries
// along with associated state changes after running a block on top.
//
// 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 {
// Immutables
root common.Hash // Root hash to which this layer diff belongs to
id uint64 // Corresponding state id
block uint64 // Associated block number
nodes map[common.Hash]map[string]*trienode.Node // Cached trie nodes indexed by owner and path
states *triestate.Set // Associated state change set for building history
memory uint64 // Approximate guess as to how much memory we use
parent layer // Parent layer modified by this one, never nil, **can be changed**
lock sync.RWMutex // Lock used to protect parent
}
// newDiffLayer creates a new diff layer on top of an existing layer.
func newDiffLayer(parent layer, root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer {
var (
size int64
count int
)
dl := &diffLayer{
root: root,
id: id,
block: block,
nodes: nodes,
states: states,
parent: parent,
}
for _, subset := range nodes {
for path, n := range subset {
dl.memory += uint64(n.Size() + len(path))
size += int64(len(n.Blob) + len(path))
}
count += len(subset)
}
if states != nil {
dl.memory += uint64(states.Size())
}
dirtyWriteMeter.Mark(size)
diffLayerNodesMeter.Mark(int64(count))
diffLayerBytesMeter.Mark(int64(dl.memory))
log.Debug("Created new diff layer", "id", id, "block", block, "nodes", count, "size", common.StorageSize(dl.memory))
return dl
}
// rootHash implements the layer interface, returning the root hash of
// corresponding state.
func (dl *diffLayer) rootHash() common.Hash {
return dl.root
}
// stateID implements the layer interface, returning the state id of the layer.
func (dl *diffLayer) stateID() uint64 {
return dl.id
}
// parentLayer implements the layer interface, returning the subsequent
// layer of the diff layer.
func (dl *diffLayer) parentLayer() layer {
dl.lock.RLock()
defer dl.lock.RUnlock()
return dl.parent
}
// node retrieves the node with provided node information. It's the internal
// version of Node function with additional accessed layer tracked. No error
// will be returned if node is not found.
func (dl *diffLayer) node(owner common.Hash, path []byte, hash common.Hash, depth int) ([]byte, error) {
// Hold the lock, ensure the parent won't be changed during the
// state accessing.
dl.lock.RLock()
defer dl.lock.RUnlock()
// If the trie node is known locally, return it
subset, ok := dl.nodes[owner]
if ok {
n, ok := subset[string(path)]
if ok {
// If the trie node is not hash matched, or marked as removed,
// bubble up an error here. It shouldn't happen at all.
if n.Hash != hash {
dirtyFalseMeter.Mark(1)
log.Error("Unexpected trie node in diff layer", "owner", owner, "path", path, "expect", hash, "got", n.Hash)
return nil, newUnexpectedNodeError("diff", hash, n.Hash, owner, path, n.Blob)
}
dirtyHitMeter.Mark(1)
dirtyNodeHitDepthHist.Update(int64(depth))
dirtyReadMeter.Mark(int64(len(n.Blob)))
return n.Blob, nil
}
}
// Trie node unknown to this layer, resolve from parent
if diff, ok := dl.parent.(*diffLayer); ok {
return diff.node(owner, path, hash, depth+1)
}
// Failed to resolve through diff layers, fallback to disk layer
return dl.parent.Node(owner, path, hash)
}
// Node implements the layer interface, retrieving the trie node blob with the
// provided node information. No error will be returned if the node is not found.
func (dl *diffLayer) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
return dl.node(owner, path, hash, 0)
}
// update implements the layer interface, creating a new layer on top of the
// existing layer tree with the specified data items.
func (dl *diffLayer) update(root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer {
return newDiffLayer(dl, root, id, block, nodes, states)
}
// persist flushes the diff layer and all its parent layers to disk layer.
func (dl *diffLayer) persist(force bool) (layer, error) {
if parent, ok := dl.parentLayer().(*diffLayer); ok {
// Hold the lock to prevent any read operation until the new
// parent is linked correctly.
dl.lock.Lock()
// The merging of diff layers starts at the bottom-most layer,
// therefore we recurse down here, flattening on the way up
// (diffToDisk).
result, err := parent.persist(force)
if err != nil {
dl.lock.Unlock()
return nil, err
}
dl.parent = result
dl.lock.Unlock()
}
return diffToDisk(dl, force)
}
// diffToDisk merges a bottom-most diff into the persistent disk layer underneath
// it. The method will panic if called onto a non-bottom-most diff layer.
func diffToDisk(layer *diffLayer, force bool) (layer, error) {
disk, ok := layer.parentLayer().(*diskLayer)
if !ok {
panic(fmt.Sprintf("unknown layer type: %T", layer.parentLayer()))
}
return disk.commit(layer, force)
}
trie/triedb/pathdb/difflayer_test.go
-170
View File
@@ -1,170 +0,0 @@
// 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 pathdb
import (
"bytes"
"testing"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/trie/testutil"
"github.com/ethereum/go-ethereum/trie/trienode"
)
func emptyLayer() *diskLayer {
return &diskLayer{
db: New(rawdb.NewMemoryDatabase(), nil),
buffer: newNodeBuffer(DefaultBufferSize, nil, 0),
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkSearch128Layers
// BenchmarkSearch128Layers-8 243826 4755 ns/op
func BenchmarkSearch128Layers(b *testing.B) { benchmarkSearch(b, 0, 128) }
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkSearch512Layers
// BenchmarkSearch512Layers-8 49686 24256 ns/op
func BenchmarkSearch512Layers(b *testing.B) { benchmarkSearch(b, 0, 512) }
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkSearch1Layer
// BenchmarkSearch1Layer-8 14062725 88.40 ns/op
func BenchmarkSearch1Layer(b *testing.B) { benchmarkSearch(b, 127, 128) }
func benchmarkSearch(b *testing.B, depth int, total int) {
var (
npath []byte
nhash common.Hash
nblob []byte
)
// First, we set up 128 diff layers, with 3K items each
fill := func(parent layer, index int) *diffLayer {
nodes := make(map[common.Hash]map[string]*trienode.Node)
nodes[common.Hash{}] = make(map[string]*trienode.Node)
for i := 0; i < 3000; i++ {
var (
path = testutil.RandBytes(32)
node = testutil.RandomNode()
)
nodes[common.Hash{}][string(path)] = trienode.New(node.Hash, node.Blob)
if npath == nil && depth == index {
npath = common.CopyBytes(path)
nblob = common.CopyBytes(node.Blob)
nhash = node.Hash
}
}
return newDiffLayer(parent, common.Hash{}, 0, 0, nodes, nil)
}
var layer layer
layer = emptyLayer()
for i := 0; i < total; i++ {
layer = fill(layer, i)
}
b.ResetTimer()
var (
have []byte
err error
)
for i := 0; i < b.N; i++ {
have, err = layer.Node(common.Hash{}, npath, nhash)
if err != nil {
b.Fatal(err)
}
}
if !bytes.Equal(have, nblob) {
b.Fatalf("have %x want %x", have, nblob)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkPersist
// BenchmarkPersist-8 10 111252975 ns/op
func BenchmarkPersist(b *testing.B) {
// First, we set up 128 diff layers, with 3K items each
fill := func(parent layer) *diffLayer {
nodes := make(map[common.Hash]map[string]*trienode.Node)
nodes[common.Hash{}] = make(map[string]*trienode.Node)
for i := 0; i < 3000; i++ {
var (
path = testutil.RandBytes(32)
node = testutil.RandomNode()
)
nodes[common.Hash{}][string(path)] = trienode.New(node.Hash, node.Blob)
}
return newDiffLayer(parent, common.Hash{}, 0, 0, nodes, nil)
}
for i := 0; i < b.N; i++ {
b.StopTimer()
var layer layer
layer = emptyLayer()
for i := 1; i < 128; i++ {
layer = fill(layer)
}
b.StartTimer()
dl, ok := layer.(*diffLayer)
if !ok {
break
}
dl.persist(false)
}
}
// BenchmarkJournal benchmarks the performance for journaling the layers.
//
// BenchmarkJournal
// BenchmarkJournal-8 10 110969279 ns/op
func BenchmarkJournal(b *testing.B) {
b.SkipNow()
// First, we set up 128 diff layers, with 3K items each
fill := func(parent layer) *diffLayer {
nodes := make(map[common.Hash]map[string]*trienode.Node)
nodes[common.Hash{}] = make(map[string]*trienode.Node)
for i := 0; i < 3000; i++ {
var (
path = testutil.RandBytes(32)
node = testutil.RandomNode()
)
nodes[common.Hash{}][string(path)] = trienode.New(node.Hash, node.Blob)
}
// TODO(rjl493456442) a non-nil state set is expected.
return newDiffLayer(parent, common.Hash{}, 0, 0, nodes, nil)
}
var layer layer
layer = emptyLayer()
for i := 0; i < 128; i++ {
layer = fill(layer)
}
b.ResetTimer()
for i := 0; i < b.N; i++ {
layer.journal(new(bytes.Buffer))
}
}
trie/triedb/pathdb/disklayer.go
-338
View File
@@ -1,338 +0,0 @@
// Copyright 2022 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 pathdb
import (
"errors"
"fmt"
"sync"
"github.com/VictoriaMetrics/fastcache"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
"golang.org/x/crypto/sha3"
)
// diskLayer is a low level persistent layer built on top of a key-value store.
type diskLayer struct {
root common.Hash // Immutable, root hash to which this layer was made for
id uint64 // Immutable, corresponding state id
db *Database // Path-based trie database
cleans *fastcache.Cache // GC friendly memory cache of clean node RLPs
buffer *nodebuffer // Node buffer to aggregate writes
stale bool // Signals that the layer became stale (state progressed)
lock sync.RWMutex // Lock used to protect stale flag
}
// newDiskLayer creates a new disk layer based on the passing arguments.
func newDiskLayer(root common.Hash, id uint64, db *Database, cleans *fastcache.Cache, buffer *nodebuffer) *diskLayer {
// Initialize a clean cache if the memory allowance is not zero
// or reuse the provided cache if it is not nil (inherited from
// the original disk layer).
if cleans == nil && db.config.CleanCacheSize != 0 {
cleans = fastcache.New(db.config.CleanCacheSize)
}
return &diskLayer{
root: root,
id: id,
db: db,
cleans: cleans,
buffer: buffer,
}
}
// root implements the layer interface, returning root hash of corresponding state.
func (dl *diskLayer) rootHash() common.Hash {
return dl.root
}
// stateID implements the layer interface, returning the state id of disk layer.
func (dl *diskLayer) stateID() uint64 {
return dl.id
}
// parent implements the layer interface, returning nil as there's no layer
// below the disk.
func (dl *diskLayer) parentLayer() layer {
return nil
}
// isStale return whether this layer has become stale (was flattened across) or if
// it's still live.
func (dl *diskLayer) isStale() bool {
dl.lock.RLock()
defer dl.lock.RUnlock()
return dl.stale
}
// markStale sets the stale flag as true.
func (dl *diskLayer) markStale() {
dl.lock.Lock()
defer dl.lock.Unlock()
if dl.stale {
panic("triedb disk layer is stale") // we've committed into the same base from two children, boom
}
dl.stale = true
}
// Node implements the layer interface, retrieving the trie node with the
// provided node info. No error will be returned if the node is not found.
func (dl *diskLayer) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
dl.lock.RLock()
defer dl.lock.RUnlock()
if dl.stale {
return nil, errSnapshotStale
}
// Try to retrieve the trie node from the not-yet-written
// node buffer first. Note the buffer is lock free since
// it's impossible to mutate the buffer before tagging the
// layer as stale.
n, err := dl.buffer.node(owner, path, hash)
if err != nil {
return nil, err
}
if n != nil {
dirtyHitMeter.Mark(1)
dirtyReadMeter.Mark(int64(len(n.Blob)))
return n.Blob, nil
}
dirtyMissMeter.Mark(1)
// Try to retrieve the trie node from the clean memory cache
key := cacheKey(owner, path)
if dl.cleans != nil {
if blob := dl.cleans.Get(nil, key); len(blob) > 0 {
h := newHasher()
defer h.release()
got := h.hash(blob)
if got == hash {
cleanHitMeter.Mark(1)
cleanReadMeter.Mark(int64(len(blob)))
return blob, nil
}
cleanFalseMeter.Mark(1)
log.Error("Unexpected trie node in clean cache", "owner", owner, "path", path, "expect", hash, "got", got)
}
cleanMissMeter.Mark(1)
}
// Try to retrieve the trie node from the disk.
var (
nBlob []byte
nHash common.Hash
)
if owner == (common.Hash{}) {
nBlob, nHash = rawdb.ReadAccountTrieNode(dl.db.diskdb, path)
} else {
nBlob, nHash = rawdb.ReadStorageTrieNode(dl.db.diskdb, owner, path)
}
if nHash != hash {
diskFalseMeter.Mark(1)
log.Error("Unexpected trie node in disk", "owner", owner, "path", path, "expect", hash, "got", nHash)
return nil, newUnexpectedNodeError("disk", hash, nHash, owner, path, nBlob)
}
if dl.cleans != nil && len(nBlob) > 0 {
dl.cleans.Set(key, nBlob)
cleanWriteMeter.Mark(int64(len(nBlob)))
}
return nBlob, nil
}
// update implements the layer interface, returning a new diff layer on top
// with the given state set.
func (dl *diskLayer) update(root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer {
return newDiffLayer(dl, root, id, block, nodes, states)
}
// commit merges the given bottom-most diff layer into the node buffer
// and returns a newly constructed disk layer. Note the current disk
// layer must be tagged as stale first to prevent re-access.
func (dl *diskLayer) commit(bottom *diffLayer, force bool) (*diskLayer, error) {
dl.lock.Lock()
defer dl.lock.Unlock()
// Construct and store the state history first. If crash happens after storing
// the state history but without flushing the corresponding states(journal),
// the stored state history will be truncated from head in the next restart.
var (
overflow bool
oldest uint64
)
if dl.db.freezer != nil {
err := writeHistory(dl.db.freezer, bottom)
if err != nil {
return nil, err
}
// Determine if the persisted history object has exceeded the configured
// limitation, set the overflow as true if so.
tail, err := dl.db.freezer.Tail()
if err != nil {
return nil, err
}
limit := dl.db.config.StateHistory
if limit != 0 && bottom.stateID()-tail > limit {
overflow = true
oldest = bottom.stateID() - limit + 1 // track the id of history **after truncation**
}
}
// Mark the diskLayer as stale before applying any mutations on top.
dl.stale = true
// Store the root->id lookup afterwards. All stored lookups are identified
// by the **unique** state root. It's impossible that in the same chain
// blocks are not adjacent but have the same root.
if dl.id == 0 {
rawdb.WriteStateID(dl.db.diskdb, dl.root, 0)
}
rawdb.WriteStateID(dl.db.diskdb, bottom.rootHash(), bottom.stateID())
// Construct a new disk layer by merging the nodes from the provided diff
// layer, and flush the content in disk layer if there are too many nodes
// cached. The clean cache is inherited from the original disk layer.
ndl := newDiskLayer(bottom.root, bottom.stateID(), dl.db, dl.cleans, dl.buffer.commit(bottom.nodes))
// In a unique scenario where the ID of the oldest history object (after tail
// truncation) surpasses the persisted state ID, we take the necessary action
// of forcibly committing the cached dirty nodes to ensure that the persisted
// state ID remains higher.
if !force && rawdb.ReadPersistentStateID(dl.db.diskdb) < oldest {
force = true
}
if err := ndl.buffer.flush(ndl.db.diskdb, ndl.cleans, ndl.id, force); err != nil {
return nil, err
}
// To remove outdated history objects from the end, we set the 'tail' parameter
// to 'oldest-1' due to the offset between the freezer index and the history ID.
if overflow {
pruned, err := truncateFromTail(ndl.db.diskdb, ndl.db.freezer, oldest-1)
if err != nil {
return nil, err
}
log.Debug("Pruned state history", "items", pruned, "tailid", oldest)
}
return ndl, nil
}
// revert applies the given state history and return a reverted disk layer.
func (dl *diskLayer) revert(h *history, loader triestate.TrieLoader) (*diskLayer, error) {
if h.meta.root != dl.rootHash() {
return nil, errUnexpectedHistory
}
// Reject if the provided state history is incomplete. It's due to
// a large construct SELF-DESTRUCT which can't be handled because
// of memory limitation.
if len(h.meta.incomplete) > 0 {
return nil, errors.New("incomplete state history")
}
if dl.id == 0 {
return nil, fmt.Errorf("%w: zero state id", errStateUnrecoverable)
}
// Apply the reverse state changes upon the current state. This must
// be done before holding the lock in order to access state in "this"
// layer.
nodes, err := triestate.Apply(h.meta.parent, h.meta.root, h.accounts, h.storages, loader)
if err != nil {
return nil, err
}
// Mark the diskLayer as stale before applying any mutations on top.
dl.lock.Lock()
defer dl.lock.Unlock()
dl.stale = true
// State change may be applied to node buffer, or the persistent
// state, depends on if node buffer is empty or not. If the node
// buffer is not empty, it means that the state transition that
// needs to be reverted is not yet flushed and cached in node
// buffer, otherwise, manipulate persistent state directly.
if !dl.buffer.empty() {
err := dl.buffer.revert(dl.db.diskdb, nodes)
if err != nil {
return nil, err
}
} else {
batch := dl.db.diskdb.NewBatch()
writeNodes(batch, nodes, dl.cleans)
rawdb.WritePersistentStateID(batch, dl.id-1)
if err := batch.Write(); err != nil {
log.Crit("Failed to write states", "err", err)
}
}
return newDiskLayer(h.meta.parent, dl.id-1, dl.db, dl.cleans, dl.buffer), nil
}
// setBufferSize sets the node buffer size to the provided value.
func (dl *diskLayer) setBufferSize(size int) error {
dl.lock.RLock()
defer dl.lock.RUnlock()
if dl.stale {
return errSnapshotStale
}
return dl.buffer.setSize(size, dl.db.diskdb, dl.cleans, dl.id)
}
// size returns the approximate size of cached nodes in the disk layer.
func (dl *diskLayer) size() common.StorageSize {
dl.lock.RLock()
defer dl.lock.RUnlock()
if dl.stale {
return 0
}
return common.StorageSize(dl.buffer.size)
}
// resetCache releases the memory held by clean cache to prevent memory leak.
func (dl *diskLayer) resetCache() {
dl.lock.RLock()
defer dl.lock.RUnlock()
// Stale disk layer loses the ownership of clean cache.
if dl.stale {
return
}
if dl.cleans != nil {
dl.cleans.Reset()
}
}
// hasher is used to compute the sha256 hash of the provided data.
type hasher struct{ sha crypto.KeccakState }
var hasherPool = sync.Pool{
New: func() interface{} { return &hasher{sha: sha3.NewLegacyKeccak256().(crypto.KeccakState)} },
}
func newHasher() *hasher {
return hasherPool.Get().(*hasher)
}
func (h *hasher) hash(data []byte) common.Hash {
return crypto.HashData(h.sha, data)
}
func (h *hasher) release() {
hasherPool.Put(h)
}
trie/triedb/pathdb/errors.go
-60
View File
@@ -1,60 +0,0 @@
// Copyright 2023 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 pathdb
import (
"errors"
"fmt"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/common/hexutil"
)
var (
// errDatabaseReadOnly is returned if the database is opened in read only mode
// to prevent any mutation.
errDatabaseReadOnly = errors.New("read only")
// errDatabaseWaitSync is returned if the initial state sync is not completed
// yet and database is disabled to prevent accessing state.
errDatabaseWaitSync = errors.New("waiting for sync")
// errSnapshotStale is returned from data accessors if the underlying layer
// layer had been invalidated due to the chain progressing forward far enough
// to not maintain the layer's original state.
errSnapshotStale = errors.New("layer stale")
// errUnexpectedHistory is returned if an unmatched state history is applied
// to the database for state rollback.
errUnexpectedHistory = errors.New("unexpected state history")
// errStateUnrecoverable is returned if state is required to be reverted to
// a destination without associated state history available.
errStateUnrecoverable = errors.New("state is unrecoverable")
// errUnexpectedNode is returned if the requested node with specified path is
// not hash matched with expectation.
errUnexpectedNode = errors.New("unexpected node")
)
func newUnexpectedNodeError(loc string, expHash common.Hash, gotHash common.Hash, owner common.Hash, path []byte, blob []byte) error {
blobHex := "nil"
if len(blob) > 0 {
blobHex = hexutil.Encode(blob)
}
return fmt.Errorf("%w, loc: %s, node: (%x %v), %x!=%x, blob: %s", errUnexpectedNode, loc, owner, path, expHash, gotHash, blobHex)
}
trie/triedb/pathdb/history.go
-649
View File
@@ -1,649 +0,0 @@
// Copyright 2022 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 pathdb
import (
"bytes"
"encoding/binary"
"errors"
"fmt"
"time"
"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/trie/triestate"
"golang.org/x/exp/slices"
)
// State history records the state changes involved in executing a block. The
// state can be reverted to the previous version by applying the associated
// history object (state reverse diff). State history objects are kept to
// guarantee that the system can perform state rollbacks in case of deep reorg.
//
// Each state transition will generate a state history object. Note that not
// every block has a corresponding state history object. If a block performs
// no state changes whatsoever, no state is created for it. Each state history
// will have a sequentially increasing number acting as its unique identifier.
//
// The state history is written to disk (ancient store) when the corresponding
// diff layer is merged into the disk layer. At the same time, system can prune
// the oldest histories according to config.
//
// Disk State
// ^
// |
// +------------+ +---------+ +---------+ +---------+
// | Init State |---->| State 1 |---->| ... |---->| State n |
// +------------+ +---------+ +---------+ +---------+
//
// +-----------+ +------+ +-----------+
// | History 1 |----> | ... |---->| History n |
// +-----------+ +------+ +-----------+
//
// # Rollback
//
// If the system wants to roll back to a previous state n, it needs to ensure
// all history objects from n+1 up to the current disk layer are existent. The
// history objects are applied to the state in reverse order, starting from the
// current disk layer.
const (
accountIndexSize = common.AddressLength + 13 // The length of encoded account index
slotIndexSize = common.HashLength + 5 // The length of encoded slot index
historyMetaSize = 9 + 2*common.HashLength // The length of fixed size part of meta object
stateHistoryVersion = uint8(0) // initial version of state history structure.
)
// Each state history entry is consisted of five elements:
//
// # metadata
// This object contains a few meta fields, such as the associated state root,
// block number, version tag and so on. This object may contain an extra
// accountHash list which means the storage changes belong to these accounts
// are not complete due to large contract destruction. The incomplete history
// can not be used for rollback and serving archive state request.
//
// # account index
// This object contains some index information of account. For example, offset
// and length indicate the location of the data belonging to the account. Besides,
// storageOffset and storageSlots indicate the storage modification location
// belonging to the account.
//
// The size of each account index is *fixed*, and all indexes are sorted
// lexicographically. Thus binary search can be performed to quickly locate a
// specific account.
//
// # account data
// Account data is a concatenated byte stream composed of all account data.
// The account data can be solved by the offset and length info indicated
// by corresponding account index.
//
// fixed size
// ^ ^
// / \
// +-----------------+-----------------+----------------+-----------------+
// | Account index 1 | Account index 2 | ... | Account index N |
// +-----------------+-----------------+----------------+-----------------+
// |
// | length
// offset |----------------+
// v v
// +----------------+----------------+----------------+----------------+
// | Account data 1 | Account data 2 | ... | Account data N |
// +----------------+----------------+----------------+----------------+
//
// # storage index
// This object is similar with account index. It's also fixed size and contains
// the location info of storage slot data.
//
// # storage data
// Storage data is a concatenated byte stream composed of all storage slot data.
// The storage slot data can be solved by the location info indicated by
// corresponding account index and storage slot index.
//
// fixed size
// ^ ^
// / \
// +-----------------+-----------------+----------------+-----------------+
// | Account index 1 | Account index 2 | ... | Account index N |
// +-----------------+-----------------+----------------+-----------------+
// |
// | storage slots
// storage offset |-----------------------------------------------------+
// v v
// +-----------------+-----------------+-----------------+
// | storage index 1 | storage index 2 | storage index 3 |
// +-----------------+-----------------+-----------------+
// | length
// offset |-------------+
// v v
// +-------------+
// | slot data 1 |
// +-------------+
// accountIndex describes the metadata belonging to an account.
type accountIndex struct {
address common.Address // The address of account
length uint8 // The length of account data, size limited by 255
offset uint32 // The offset of item in account data table
storageOffset uint32 // The offset of storage index in storage index table
storageSlots uint32 // The number of mutated storage slots belonging to the account
}
// encode packs account index into byte stream.
func (i *accountIndex) encode() []byte {
var buf [accountIndexSize]byte
copy(buf[:], i.address.Bytes())
buf[common.AddressLength] = i.length
binary.BigEndian.PutUint32(buf[common.AddressLength+1:], i.offset)
binary.BigEndian.PutUint32(buf[common.AddressLength+5:], i.storageOffset)
binary.BigEndian.PutUint32(buf[common.AddressLength+9:], i.storageSlots)
return buf[:]
}
// decode unpacks account index from byte stream.
func (i *accountIndex) decode(blob []byte) {
i.address = common.BytesToAddress(blob[:common.AddressLength])
i.length = blob[common.AddressLength]
i.offset = binary.BigEndian.Uint32(blob[common.AddressLength+1:])
i.storageOffset = binary.BigEndian.Uint32(blob[common.AddressLength+5:])
i.storageSlots = binary.BigEndian.Uint32(blob[common.AddressLength+9:])
}
// slotIndex describes the metadata belonging to a storage slot.
type slotIndex struct {
hash common.Hash // The hash of slot key
length uint8 // The length of storage slot, up to 32 bytes defined in protocol
offset uint32 // The offset of item in storage slot data table
}
// encode packs slot index into byte stream.
func (i *slotIndex) encode() []byte {
var buf [slotIndexSize]byte
copy(buf[:common.HashLength], i.hash.Bytes())
buf[common.HashLength] = i.length
binary.BigEndian.PutUint32(buf[common.HashLength+1:], i.offset)
return buf[:]
}
// decode unpack slot index from the byte stream.
func (i *slotIndex) decode(blob []byte) {
i.hash = common.BytesToHash(blob[:common.HashLength])
i.length = blob[common.HashLength]
i.offset = binary.BigEndian.Uint32(blob[common.HashLength+1:])
}
// meta describes the meta data of state history object.
type meta struct {
version uint8 // version tag of history object
parent common.Hash // prev-state root before the state transition
root common.Hash // post-state root after the state transition
block uint64 // associated block number
incomplete []common.Address // list of address whose storage set is incomplete
}
// encode packs the meta object into byte stream.
func (m *meta) encode() []byte {
buf := make([]byte, historyMetaSize+len(m.incomplete)*common.AddressLength)
buf[0] = m.version
copy(buf[1:1+common.HashLength], m.parent.Bytes())
copy(buf[1+common.HashLength:1+2*common.HashLength], m.root.Bytes())
binary.BigEndian.PutUint64(buf[1+2*common.HashLength:historyMetaSize], m.block)
for i, h := range m.incomplete {
copy(buf[i*common.AddressLength+historyMetaSize:], h.Bytes())
}
return buf[:]
}
// decode unpacks the meta object from byte stream.
func (m *meta) decode(blob []byte) error {
if len(blob) < 1 {
return fmt.Errorf("no version tag")
}
switch blob[0] {
case stateHistoryVersion:
if len(blob) < historyMetaSize {
return fmt.Errorf("invalid state history meta, len: %d", len(blob))
}
if (len(blob)-historyMetaSize)%common.AddressLength != 0 {
return fmt.Errorf("corrupted state history meta, len: %d", len(blob))
}
m.version = blob[0]
m.parent = common.BytesToHash(blob[1 : 1+common.HashLength])
m.root = common.BytesToHash(blob[1+common.HashLength : 1+2*common.HashLength])
m.block = binary.BigEndian.Uint64(blob[1+2*common.HashLength : historyMetaSize])
for pos := historyMetaSize; pos < len(blob); {
m.incomplete = append(m.incomplete, common.BytesToAddress(blob[pos:pos+common.AddressLength]))
pos += common.AddressLength
}
return nil
default:
return fmt.Errorf("unknown version %d", blob[0])
}
}
// history represents a set of state changes belong to a block along with
// the metadata including the state roots involved in the state transition.
// State history objects in disk are linked with each other by a unique id
// (8-bytes integer), the oldest state history object can be pruned on demand
// in order to control the storage size.
type history struct {
meta *meta // Meta data of history
accounts map[common.Address][]byte // Account data keyed by its address hash
accountList []common.Address // Sorted account hash list
storages map[common.Address]map[common.Hash][]byte // Storage data keyed by its address hash and slot hash
storageList map[common.Address][]common.Hash // Sorted slot hash list
}
// newHistory constructs the state history object with provided state change set.
func newHistory(root common.Hash, parent common.Hash, block uint64, states *triestate.Set) *history {
var (
accountList []common.Address
storageList = make(map[common.Address][]common.Hash)
incomplete []common.Address
)
for addr := range states.Accounts {
accountList = append(accountList, addr)
}
slices.SortFunc(accountList, common.Address.Cmp)
for addr, slots := range states.Storages {
slist := make([]common.Hash, 0, len(slots))
for slotHash := range slots {
slist = append(slist, slotHash)
}
slices.SortFunc(slist, common.Hash.Cmp)
storageList[addr] = slist
}
for addr := range states.Incomplete {
incomplete = append(incomplete, addr)
}
slices.SortFunc(incomplete, common.Address.Cmp)
return &history{
meta: &meta{
version: stateHistoryVersion,
parent: parent,
root: root,
block: block,
incomplete: incomplete,
},
accounts: states.Accounts,
accountList: accountList,
storages: states.Storages,
storageList: storageList,
}
}
// encode serializes the state history and returns four byte streams represent
// concatenated account/storage data, account/storage indexes respectively.
func (h *history) encode() ([]byte, []byte, []byte, []byte) {
var (
slotNumber uint32 // the number of processed slots
accountData []byte // the buffer for concatenated account data
storageData []byte // the buffer for concatenated storage data
accountIndexes []byte // the buffer for concatenated account index
storageIndexes []byte // the buffer for concatenated storage index
)
for _, addr := range h.accountList {
accIndex := accountIndex{
address: addr,
length: uint8(len(h.accounts[addr])),
offset: uint32(len(accountData)),
}
slots, exist := h.storages[addr]
if exist {
// Encode storage slots in order
for _, slotHash := range h.storageList[addr] {
sIndex := slotIndex{
hash: slotHash,
length: uint8(len(slots[slotHash])),
offset: uint32(len(storageData)),
}
storageData = append(storageData, slots[slotHash]...)
storageIndexes = append(storageIndexes, sIndex.encode()...)
}
// Fill up the storage meta in account index
accIndex.storageOffset = slotNumber
accIndex.storageSlots = uint32(len(slots))
slotNumber += uint32(len(slots))
}
accountData = append(accountData, h.accounts[addr]...)
accountIndexes = append(accountIndexes, accIndex.encode()...)
}
return accountData, storageData, accountIndexes, storageIndexes
}
// decoder wraps the byte streams for decoding with extra meta fields.
type decoder struct {
accountData []byte // the buffer for concatenated account data
storageData []byte // the buffer for concatenated storage data
accountIndexes []byte // the buffer for concatenated account index
storageIndexes []byte // the buffer for concatenated storage index
lastAccount *common.Address // the address of last resolved account
lastAccountRead uint32 // the read-cursor position of account data
lastSlotIndexRead uint32 // the read-cursor position of storage slot index
lastSlotDataRead uint32 // the read-cursor position of storage slot data
}
// verify validates the provided byte streams for decoding state history. A few
// checks will be performed to quickly detect data corruption. The byte stream
// is regarded as corrupted if:
//
// - account indexes buffer is empty(empty state set is invalid)
// - account indexes/storage indexer buffer is not aligned
//
// note, these situations are allowed:
//
// - empty account data: all accounts were not present
// - empty storage set: no slots are modified
func (r *decoder) verify() error {
if len(r.accountIndexes)%accountIndexSize != 0 || len(r.accountIndexes) == 0 {
return fmt.Errorf("invalid account index, len: %d", len(r.accountIndexes))
}
if len(r.storageIndexes)%slotIndexSize != 0 {
return fmt.Errorf("invalid storage index, len: %d", len(r.storageIndexes))
}
return nil
}
// readAccount parses the account from the byte stream with specified position.
func (r *decoder) readAccount(pos int) (accountIndex, []byte, error) {
// Decode account index from the index byte stream.
var index accountIndex
if (pos+1)*accountIndexSize > len(r.accountIndexes) {
return accountIndex{}, nil, errors.New("account data buffer is corrupted")
}
index.decode(r.accountIndexes[pos*accountIndexSize : (pos+1)*accountIndexSize])
// Perform validation before parsing account data, ensure
// - account is sorted in order in byte stream
// - account data is strictly encoded with no gap inside
// - account data is not out-of-slice
if r.lastAccount != nil { // zero address is possible
if bytes.Compare(r.lastAccount.Bytes(), index.address.Bytes()) >= 0 {
return accountIndex{}, nil, errors.New("account is not in order")
}
}
if index.offset != r.lastAccountRead {
return accountIndex{}, nil, errors.New("account data buffer is gaped")
}
last := index.offset + uint32(index.length)
if uint32(len(r.accountData)) < last {
return accountIndex{}, nil, errors.New("account data buffer is corrupted")
}
data := r.accountData[index.offset:last]
r.lastAccount = &index.address
r.lastAccountRead = last
return index, data, nil
}
// readStorage parses the storage slots from the byte stream with specified account.
func (r *decoder) readStorage(accIndex accountIndex) ([]common.Hash, map[common.Hash][]byte, error) {
var (
last common.Hash
list []common.Hash
storage = make(map[common.Hash][]byte)
)
for j := 0; j < int(accIndex.storageSlots); j++ {
var (
index slotIndex
start = (accIndex.storageOffset + uint32(j)) * uint32(slotIndexSize)
end = (accIndex.storageOffset + uint32(j+1)) * uint32(slotIndexSize)
)
// Perform validation before parsing storage slot data, ensure
// - slot index is not out-of-slice
// - slot data is not out-of-slice
// - slot is sorted in order in byte stream
// - slot indexes is strictly encoded with no gap inside
// - slot data is strictly encoded with no gap inside
if start != r.lastSlotIndexRead {
return nil, nil, errors.New("storage index buffer is gapped")
}
if uint32(len(r.storageIndexes)) < end {
return nil, nil, errors.New("storage index buffer is corrupted")
}
index.decode(r.storageIndexes[start:end])
if bytes.Compare(last.Bytes(), index.hash.Bytes()) >= 0 {
return nil, nil, errors.New("storage slot is not in order")
}
if index.offset != r.lastSlotDataRead {
return nil, nil, errors.New("storage data buffer is gapped")
}
sEnd := index.offset + uint32(index.length)
if uint32(len(r.storageData)) < sEnd {
return nil, nil, errors.New("storage data buffer is corrupted")
}
storage[index.hash] = r.storageData[r.lastSlotDataRead:sEnd]
list = append(list, index.hash)
last = index.hash
r.lastSlotIndexRead = end
r.lastSlotDataRead = sEnd
}
return list, storage, nil
}
// decode deserializes the account and storage data from the provided byte stream.
func (h *history) decode(accountData, storageData, accountIndexes, storageIndexes []byte) error {
var (
accounts = make(map[common.Address][]byte)
storages = make(map[common.Address]map[common.Hash][]byte)
accountList []common.Address
storageList = make(map[common.Address][]common.Hash)
r = &decoder{
accountData: accountData,
storageData: storageData,
accountIndexes: accountIndexes,
storageIndexes: storageIndexes,
}
)
if err := r.verify(); err != nil {
return err
}
for i := 0; i < len(accountIndexes)/accountIndexSize; i++ {
// Resolve account first
accIndex, accData, err := r.readAccount(i)
if err != nil {
return err
}
accounts[accIndex.address] = accData
accountList = append(accountList, accIndex.address)
// Resolve storage slots
slotList, slotData, err := r.readStorage(accIndex)
if err != nil {
return err
}
if len(slotList) > 0 {
storageList[accIndex.address] = slotList
storages[accIndex.address] = slotData
}
}
h.accounts = accounts
h.accountList = accountList
h.storages = storages
h.storageList = storageList
return nil
}
// readHistory reads and decodes the state history object by the given id.
func readHistory(freezer *rawdb.ResettableFreezer, id uint64) (*history, error) {
blob := rawdb.ReadStateHistoryMeta(freezer, id)
if len(blob) == 0 {
return nil, fmt.Errorf("state history not found %d", id)
}
var m meta
if err := m.decode(blob); err != nil {
return nil, err
}
var (
dec = history{meta: &m}
accountData = rawdb.ReadStateAccountHistory(freezer, id)
storageData = rawdb.ReadStateStorageHistory(freezer, id)
accountIndexes = rawdb.ReadStateAccountIndex(freezer, id)
storageIndexes = rawdb.ReadStateStorageIndex(freezer, id)
)
if err := dec.decode(accountData, storageData, accountIndexes, storageIndexes); err != nil {
return nil, err
}
return &dec, nil
}
// writeHistory persists the state history with the provided state set.
func writeHistory(freezer *rawdb.ResettableFreezer, dl *diffLayer) error {
// Short circuit if state set is not available.
if dl.states == nil {
return errors.New("state change set is not available")
}
var (
start = time.Now()
history = newHistory(dl.rootHash(), dl.parentLayer().rootHash(), dl.block, dl.states)
)
accountData, storageData, accountIndex, storageIndex := history.encode()
dataSize := common.StorageSize(len(accountData) + len(storageData))
indexSize := common.StorageSize(len(accountIndex) + len(storageIndex))
// Write history data into five freezer table respectively.
rawdb.WriteStateHistory(freezer, dl.stateID(), history.meta.encode(), accountIndex, storageIndex, accountData, storageData)
historyDataBytesMeter.Mark(int64(dataSize))
historyIndexBytesMeter.Mark(int64(indexSize))
historyBuildTimeMeter.UpdateSince(start)
log.Debug("Stored state history", "id", dl.stateID(), "block", dl.block, "data", dataSize, "index", indexSize, "elapsed", common.PrettyDuration(time.Since(start)))
return nil
}
// checkHistories retrieves a batch of meta objects with the specified range
// and performs the callback on each item.
func checkHistories(freezer *rawdb.ResettableFreezer, start, count uint64, check func(*meta) error) error {
for count > 0 {
number := count
if number > 10000 {
number = 10000 // split the big read into small chunks
}
blobs, err := rawdb.ReadStateHistoryMetaList(freezer, start, number)
if err != nil {
return err
}
for _, blob := range blobs {
var dec meta
if err := dec.decode(blob); err != nil {
return err
}
if err := check(&dec); err != nil {
return err
}
}
count -= uint64(len(blobs))
start += uint64(len(blobs))
}
return nil
}
// truncateFromHead removes the extra state histories from the head with the given
// parameters. It returns the number of items removed from the head.
func truncateFromHead(db ethdb.Batcher, freezer *rawdb.ResettableFreezer, nhead uint64) (int, error) {
ohead, err := freezer.Ancients()
if err != nil {
return 0, err
}
otail, err := freezer.Tail()
if err != nil {
return 0, err
}
// Ensure that the truncation target falls within the specified range.
if ohead < nhead || nhead < otail {
return 0, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", otail, ohead, nhead)
}
// Short circuit if nothing to truncate.
if ohead == nhead {
return 0, nil
}
// Load the meta objects in range [nhead+1, ohead]
blobs, err := rawdb.ReadStateHistoryMetaList(freezer, nhead+1, ohead-nhead)
if err != nil {
return 0, err
}
batch := db.NewBatch()
for _, blob := range blobs {
var m meta
if err := m.decode(blob); err != nil {
return 0, err
}
rawdb.DeleteStateID(batch, m.root)
}
if err := batch.Write(); err != nil {
return 0, err
}
ohead, err = freezer.TruncateHead(nhead)
if err != nil {
return 0, err
}
return int(ohead - nhead), nil
}
// truncateFromTail removes the extra state histories from the tail with the given
// parameters. It returns the number of items removed from the tail.
func truncateFromTail(db ethdb.Batcher, freezer *rawdb.ResettableFreezer, ntail uint64) (int, error) {
ohead, err := freezer.Ancients()
if err != nil {
return 0, err
}
otail, err := freezer.Tail()
if err != nil {
return 0, err
}
// Ensure that the truncation target falls within the specified range.
if otail > ntail || ntail > ohead {
return 0, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", otail, ohead, ntail)
}
// Short circuit if nothing to truncate.
if otail == ntail {
return 0, nil
}
// Load the meta objects in range [otail+1, ntail]
blobs, err := rawdb.ReadStateHistoryMetaList(freezer, otail+1, ntail-otail)
if err != nil {
return 0, err
}
batch := db.NewBatch()
for _, blob := range blobs {
var m meta
if err := m.decode(blob); err != nil {
return 0, err
}
rawdb.DeleteStateID(batch, m.root)
}
if err := batch.Write(); err != nil {
return 0, err
}
otail, err = freezer.TruncateTail(ntail)
if err != nil {
return 0, err
}
return int(ntail - otail), nil
}
trie/triedb/pathdb/history_test.go
-334
View File
@@ -1,334 +0,0 @@
// Copyright 2022 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 pathdb
import (
"bytes"
"fmt"
"reflect"
"testing"
"github.com/ethereum/go-ethereum/common"
"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/rlp"
"github.com/ethereum/go-ethereum/trie/testutil"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// randomStateSet generates a random state change set.
func randomStateSet(n int) *triestate.Set {
var (
accounts = make(map[common.Address][]byte)
storages = make(map[common.Address]map[common.Hash][]byte)
)
for i := 0; i < n; i++ {
addr := testutil.RandomAddress()
storages[addr] = make(map[common.Hash][]byte)
for j := 0; j < 3; j++ {
v, _ := rlp.EncodeToBytes(common.TrimLeftZeroes(testutil.RandBytes(32)))
storages[addr][testutil.RandomHash()] = v
}
account := generateAccount(types.EmptyRootHash)
accounts[addr] = types.SlimAccountRLP(account)
}
return triestate.New(accounts, storages, nil)
}
func makeHistory() *history {
return newHistory(testutil.RandomHash(), types.EmptyRootHash, 0, randomStateSet(3))
}
func makeHistories(n int) []*history {
var (
parent = types.EmptyRootHash
result []*history
)
for i := 0; i < n; i++ {
root := testutil.RandomHash()
h := newHistory(root, parent, uint64(i), randomStateSet(3))
parent = root
result = append(result, h)
}
return result
}
func TestEncodeDecodeHistory(t *testing.T) {
var (
m meta
dec history
obj = makeHistory()
)
// check if meta data can be correctly encode/decode
blob := obj.meta.encode()
if err := m.decode(blob); err != nil {
t.Fatalf("Failed to decode %v", err)
}
if !reflect.DeepEqual(&m, obj.meta) {
t.Fatal("meta is mismatched")
}
// check if account/storage data can be correctly encode/decode
accountData, storageData, accountIndexes, storageIndexes := obj.encode()
if err := dec.decode(accountData, storageData, accountIndexes, storageIndexes); err != nil {
t.Fatalf("Failed to decode, err: %v", err)
}
if !compareSet(dec.accounts, obj.accounts) {
t.Fatal("account data is mismatched")
}
if !compareStorages(dec.storages, obj.storages) {
t.Fatal("storage data is mismatched")
}
if !compareList(dec.accountList, obj.accountList) {
t.Fatal("account list is mismatched")
}
if !compareStorageList(dec.storageList, obj.storageList) {
t.Fatal("storage list is mismatched")
}
}
func checkHistory(t *testing.T, db ethdb.KeyValueReader, freezer *rawdb.ResettableFreezer, id uint64, root common.Hash, exist bool) {
blob := rawdb.ReadStateHistoryMeta(freezer, id)
if exist && len(blob) == 0 {
t.Fatalf("Failed to load trie history, %d", id)
}
if !exist && len(blob) != 0 {
t.Fatalf("Unexpected trie history, %d", id)
}
if exist && rawdb.ReadStateID(db, root) == nil {
t.Fatalf("Root->ID mapping is not found, %d", id)
}
if !exist && rawdb.ReadStateID(db, root) != nil {
t.Fatalf("Unexpected root->ID mapping, %d", id)
}
}
func checkHistoriesInRange(t *testing.T, db ethdb.KeyValueReader, freezer *rawdb.ResettableFreezer, from, to uint64, roots []common.Hash, exist bool) {
for i, j := from, 0; i <= to; i, j = i+1, j+1 {
checkHistory(t, db, freezer, i, roots[j], exist)
}
}
func TestTruncateHeadHistory(t *testing.T) {
var (
roots []common.Hash
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir(), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
roots = append(roots, hs[i].meta.root)
}
for size := len(hs); size > 0; size-- {
pruned, err := truncateFromHead(db, freezer, uint64(size-1))
if err != nil {
t.Fatalf("Failed to truncate from head %v", err)
}
if pruned != 1 {
t.Error("Unexpected pruned items", "want", 1, "got", pruned)
}
checkHistoriesInRange(t, db, freezer, uint64(size), uint64(10), roots[size-1:], false)
checkHistoriesInRange(t, db, freezer, uint64(1), uint64(size-1), roots[:size-1], true)
}
}
func TestTruncateTailHistory(t *testing.T) {
var (
roots []common.Hash
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir(), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
roots = append(roots, hs[i].meta.root)
}
for newTail := 1; newTail < len(hs); newTail++ {
pruned, _ := truncateFromTail(db, freezer, uint64(newTail))
if pruned != 1 {
t.Error("Unexpected pruned items", "want", 1, "got", pruned)
}
checkHistoriesInRange(t, db, freezer, uint64(1), uint64(newTail), roots[:newTail], false)
checkHistoriesInRange(t, db, freezer, uint64(newTail+1), uint64(10), roots[newTail:], true)
}
}
func TestTruncateTailHistories(t *testing.T) {
var cases = []struct {
limit uint64
expPruned int
maxPruned uint64
minUnpruned uint64
empty bool
}{
{
1, 9, 9, 10, false,
},
{
0, 10, 10, 0 /* no meaning */, true,
},
{
10, 0, 0, 1, false,
},
}
for i, c := range cases {
var (
roots []common.Hash
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir()+fmt.Sprintf("%d", i), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
roots = append(roots, hs[i].meta.root)
}
pruned, _ := truncateFromTail(db, freezer, uint64(10)-c.limit)
if pruned != c.expPruned {
t.Error("Unexpected pruned items", "want", c.expPruned, "got", pruned)
}
if c.empty {
checkHistoriesInRange(t, db, freezer, uint64(1), uint64(10), roots, false)
} else {
tail := 10 - int(c.limit)
checkHistoriesInRange(t, db, freezer, uint64(1), c.maxPruned, roots[:tail], false)
checkHistoriesInRange(t, db, freezer, c.minUnpruned, uint64(10), roots[tail:], true)
}
}
}
func TestTruncateOutOfRange(t *testing.T) {
var (
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir(), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
}
truncateFromTail(db, freezer, uint64(len(hs)/2))
// Ensure of-out-range truncations are rejected correctly.
head, _ := freezer.Ancients()
tail, _ := freezer.Tail()
cases := []struct {
mode int
target uint64
expErr error
}{
{0, head, nil}, // nothing to delete
{0, head + 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, head+1)},
{0, tail - 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, tail-1)},
{1, tail, nil}, // nothing to delete
{1, head + 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, head+1)},
{1, tail - 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, tail-1)},
}
for _, c := range cases {
var gotErr error
if c.mode == 0 {
_, gotErr = truncateFromHead(db, freezer, c.target)
} else {
_, gotErr = truncateFromTail(db, freezer, c.target)
}
if !reflect.DeepEqual(gotErr, c.expErr) {
t.Errorf("Unexpected error, want: %v, got: %v", c.expErr, gotErr)
}
}
}
// openFreezer initializes the freezer instance for storing state histories.
func openFreezer(datadir string, readOnly bool) (*rawdb.ResettableFreezer, error) {
return rawdb.NewStateFreezer(datadir, readOnly)
}
func compareSet[k comparable](a, b map[k][]byte) bool {
if len(a) != len(b) {
return false
}
for key, valA := range a {
valB, ok := b[key]
if !ok {
return false
}
if !bytes.Equal(valA, valB) {
return false
}
}
return true
}
func compareList[k comparable](a, b []k) bool {
if len(a) != len(b) {
return false
}
for i := 0; i < len(a); i++ {
if a[i] != b[i] {
return false
}
}
return true
}
func compareStorages(a, b map[common.Address]map[common.Hash][]byte) bool {
if len(a) != len(b) {
return false
}
for h, subA := range a {
subB, ok := b[h]
if !ok {
return false
}
if !compareSet(subA, subB) {
return false
}
}
return true
}
func compareStorageList(a, b map[common.Address][]common.Hash) bool {
if len(a) != len(b) {
return false
}
for h, la := range a {
lb, ok := b[h]
if !ok {
return false
}
if !compareList(la, lb) {
return false
}
}
return true
}
trie/triedb/pathdb/journal.go
-387
View File
@@ -1,387 +0,0 @@
// Copyright 2022 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 pathdb
import (
"bytes"
"errors"
"fmt"
"io"
"time"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
var (
errMissJournal = errors.New("journal not found")
errMissVersion = errors.New("version not found")
errUnexpectedVersion = errors.New("unexpected journal version")
errMissDiskRoot = errors.New("disk layer root not found")
errUnmatchedJournal = errors.New("unmatched journal")
)
const journalVersion uint64 = 0
// journalNode represents a trie node persisted in the journal.
type journalNode struct {
Path []byte // Path of the node in the trie
Blob []byte // RLP-encoded trie node blob, nil means the node is deleted
}
// journalNodes represents a list trie nodes belong to a single account
// or the main account trie.
type journalNodes struct {
Owner common.Hash
Nodes []journalNode
}
// journalAccounts represents a list accounts belong to the layer.
type journalAccounts struct {
Addresses []common.Address
Accounts [][]byte
}
// journalStorage represents a list of storage slots belong to an account.
type journalStorage struct {
Incomplete bool
Account common.Address
Hashes []common.Hash
Slots [][]byte
}
// loadJournal tries to parse the layer journal from the disk.
func (db *Database) loadJournal(diskRoot common.Hash) (layer, error) {
journal := rawdb.ReadTrieJournal(db.diskdb)
if len(journal) == 0 {
return nil, errMissJournal
}
r := rlp.NewStream(bytes.NewReader(journal), 0)
// Firstly, resolve the first element as the journal version
version, err := r.Uint64()
if err != nil {
return nil, errMissVersion
}
if version != journalVersion {
return nil, fmt.Errorf("%w want %d got %d", errUnexpectedVersion, journalVersion, version)
}
// Secondly, resolve the disk layer root, ensure it's continuous
// with disk layer. Note now we can ensure it's the layer journal
// correct version, so we expect everything can be resolved properly.
var root common.Hash
if err := r.Decode(&root); err != nil {
return nil, errMissDiskRoot
}
// The journal is not matched with persistent state, discard them.
// It can happen that geth crashes without persisting the journal.
if !bytes.Equal(root.Bytes(), diskRoot.Bytes()) {
return nil, fmt.Errorf("%w want %x got %x", errUnmatchedJournal, root, diskRoot)
}
// Load the disk layer from the journal
base, err := db.loadDiskLayer(r)
if err != nil {
return nil, err
}
// Load all the diff layers from the journal
head, err := db.loadDiffLayer(base, r)
if err != nil {
return nil, err
}
log.Debug("Loaded layer journal", "diskroot", diskRoot, "diffhead", head.rootHash())
return head, nil
}
// loadLayers loads a pre-existing state layer backed by a key-value store.
func (db *Database) loadLayers() layer {
// Retrieve the root node of persistent state.
_, root := rawdb.ReadAccountTrieNode(db.diskdb, nil)
root = types.TrieRootHash(root)
// Load the layers by resolving the journal
head, err := db.loadJournal(root)
if err == nil {
return head
}
// journal is not matched(or missing) with the persistent state, discard
// it. Display log for discarding journal, but try to avoid showing
// useless information when the db is created from scratch.
if !(root == types.EmptyRootHash && errors.Is(err, errMissJournal)) {
log.Info("Failed to load journal, discard it", "err", err)
}
// Return single layer with persistent state.
return newDiskLayer(root, rawdb.ReadPersistentStateID(db.diskdb), db, nil, newNodeBuffer(db.bufferSize, nil, 0))
}
// loadDiskLayer reads the binary blob from the layer journal, reconstructing
// a new disk layer on it.
func (db *Database) loadDiskLayer(r *rlp.Stream) (layer, error) {
// Resolve disk layer root
var root common.Hash
if err := r.Decode(&root); err != nil {
return nil, fmt.Errorf("load disk root: %v", err)
}
// Resolve the state id of disk layer, it can be different
// with the persistent id tracked in disk, the id distance
// is the number of transitions aggregated in disk layer.
var id uint64
if err := r.Decode(&id); err != nil {
return nil, fmt.Errorf("load state id: %v", err)
}
stored := rawdb.ReadPersistentStateID(db.diskdb)
if stored > id {
return nil, fmt.Errorf("invalid state id: stored %d resolved %d", stored, id)
}
// Resolve nodes cached in node buffer
var encoded []journalNodes
if err := r.Decode(&encoded); err != nil {
return nil, fmt.Errorf("load disk nodes: %v", err)
}
nodes := make(map[common.Hash]map[string]*trienode.Node)
for _, entry := range encoded {
subset := make(map[string]*trienode.Node)
for _, n := range entry.Nodes {
if len(n.Blob) > 0 {
subset[string(n.Path)] = trienode.New(crypto.Keccak256Hash(n.Blob), n.Blob)
} else {
subset[string(n.Path)] = trienode.NewDeleted()
}
}
nodes[entry.Owner] = subset
}
// Calculate the internal state transitions by id difference.
base := newDiskLayer(root, id, db, nil, newNodeBuffer(db.bufferSize, nodes, id-stored))
return base, nil
}
// loadDiffLayer reads the next sections of a layer journal, reconstructing a new
// diff and verifying that it can be linked to the requested parent.
func (db *Database) loadDiffLayer(parent layer, r *rlp.Stream) (layer, error) {
// Read the next diff journal entry
var root common.Hash
if err := r.Decode(&root); 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 root: %v", err)
}
var block uint64
if err := r.Decode(&block); err != nil {
return nil, fmt.Errorf("load block number: %v", err)
}
// Read in-memory trie nodes from journal
var encoded []journalNodes
if err := r.Decode(&encoded); err != nil {
return nil, fmt.Errorf("load diff nodes: %v", err)
}
nodes := make(map[common.Hash]map[string]*trienode.Node)
for _, entry := range encoded {
subset := make(map[string]*trienode.Node)
for _, n := range entry.Nodes {
if len(n.Blob) > 0 {
subset[string(n.Path)] = trienode.New(crypto.Keccak256Hash(n.Blob), n.Blob)
} else {
subset[string(n.Path)] = trienode.NewDeleted()
}
}
nodes[entry.Owner] = subset
}
// Read state changes from journal
var (
jaccounts journalAccounts
jstorages []journalStorage
accounts = make(map[common.Address][]byte)
storages = make(map[common.Address]map[common.Hash][]byte)
incomplete = make(map[common.Address]struct{})
)
if err := r.Decode(&jaccounts); err != nil {
return nil, fmt.Errorf("load diff accounts: %v", err)
}
for i, addr := range jaccounts.Addresses {
accounts[addr] = jaccounts.Accounts[i]
}
if err := r.Decode(&jstorages); err != nil {
return nil, fmt.Errorf("load diff storages: %v", err)
}
for _, entry := range jstorages {
set := make(map[common.Hash][]byte)
for i, h := range entry.Hashes {
if len(entry.Slots[i]) > 0 {
set[h] = entry.Slots[i]
} else {
set[h] = nil
}
}
if entry.Incomplete {
incomplete[entry.Account] = struct{}{}
}
storages[entry.Account] = set
}
return db.loadDiffLayer(newDiffLayer(parent, root, parent.stateID()+1, block, nodes, triestate.New(accounts, storages, incomplete)), r)
}
// journal implements the layer interface, marshaling the un-flushed trie nodes
// along with layer meta data into provided byte buffer.
func (dl *diskLayer) journal(w io.Writer) error {
dl.lock.RLock()
defer dl.lock.RUnlock()
// Ensure the layer didn't get stale
if dl.stale {
return errSnapshotStale
}
// Step one, write the disk root into the journal.
if err := rlp.Encode(w, dl.root); err != nil {
return err
}
// Step two, write the corresponding state id into the journal
if err := rlp.Encode(w, dl.id); err != nil {
return err
}
// Step three, write all unwritten nodes into the journal
nodes := make([]journalNodes, 0, len(dl.buffer.nodes))
for owner, subset := range dl.buffer.nodes {
entry := journalNodes{Owner: owner}
for path, node := range subset {
entry.Nodes = append(entry.Nodes, journalNode{Path: []byte(path), Blob: node.Blob})
}
nodes = append(nodes, entry)
}
if err := rlp.Encode(w, nodes); err != nil {
return err
}
log.Debug("Journaled pathdb disk layer", "root", dl.root, "nodes", len(dl.buffer.nodes))
return nil
}
// journal implements the layer interface, writing the memory layer contents
// into a buffer to be stored in the database as the layer journal.
func (dl *diffLayer) journal(w io.Writer) error {
dl.lock.RLock()
defer dl.lock.RUnlock()
// journal the parent first
if err := dl.parent.journal(w); err != nil {
return err
}
// Everything below was journaled, persist this layer too
if err := rlp.Encode(w, dl.root); err != nil {
return err
}
if err := rlp.Encode(w, dl.block); err != nil {
return err
}
// Write the accumulated trie nodes into buffer
nodes := make([]journalNodes, 0, len(dl.nodes))
for owner, subset := range dl.nodes {
entry := journalNodes{Owner: owner}
for path, node := range subset {
entry.Nodes = append(entry.Nodes, journalNode{Path: []byte(path), Blob: node.Blob})
}
nodes = append(nodes, entry)
}
if err := rlp.Encode(w, nodes); err != nil {
return err
}
// Write the accumulated state changes into buffer
var jacct journalAccounts
for addr, account := range dl.states.Accounts {
jacct.Addresses = append(jacct.Addresses, addr)
jacct.Accounts = append(jacct.Accounts, account)
}
if err := rlp.Encode(w, jacct); err != nil {
return err
}
storage := make([]journalStorage, 0, len(dl.states.Storages))
for addr, slots := range dl.states.Storages {
entry := journalStorage{Account: addr}
if _, ok := dl.states.Incomplete[addr]; ok {
entry.Incomplete = true
}
for slotHash, slot := range slots {
entry.Hashes = append(entry.Hashes, slotHash)
entry.Slots = append(entry.Slots, slot)
}
storage = append(storage, entry)
}
if err := rlp.Encode(w, storage); err != nil {
return err
}
log.Debug("Journaled pathdb diff layer", "root", dl.root, "parent", dl.parent.rootHash(), "id", dl.stateID(), "block", dl.block, "nodes", len(dl.nodes))
return nil
}
// Journal commits an entire diff hierarchy to disk into a single journal entry.
// This is meant to be used during shutdown to persist the layer without
// flattening everything down (bad for reorgs). And this function will mark the
// database as read-only to prevent all following mutation to disk.
func (db *Database) Journal(root common.Hash) error {
// Retrieve the head layer to journal from.
l := db.tree.get(root)
if l == nil {
return fmt.Errorf("triedb layer [%#x] missing", root)
}
disk := db.tree.bottom()
if l, ok := l.(*diffLayer); ok {
log.Info("Persisting dirty state to disk", "head", l.block, "root", root, "layers", l.id-disk.id+disk.buffer.layers)
} else { // disk layer only on noop runs (likely) or deep reorgs (unlikely)
log.Info("Persisting dirty state to disk", "root", root, "layers", disk.buffer.layers)
}
start := time.Now()
// Run the journaling
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the database is in read only mode.
if db.readOnly {
return errDatabaseReadOnly
}
// Firstly write out the metadata of journal
journal := new(bytes.Buffer)
if err := rlp.Encode(journal, journalVersion); err != nil {
return err
}
// The stored state in disk might be empty, convert the
// root to emptyRoot in this case.
_, diskroot := rawdb.ReadAccountTrieNode(db.diskdb, nil)
diskroot = types.TrieRootHash(diskroot)
// Secondly write out the state root in disk, ensure all layers
// on top are continuous with disk.
if err := rlp.Encode(journal, diskroot); err != nil {
return err
}
// Finally write out the journal of each layer in reverse order.
if err := l.journal(journal); err != nil {
return err
}
// Store the journal into the database and return
rawdb.WriteTrieJournal(db.diskdb, journal.Bytes())
// Set the db in read only mode to reject all following mutations
db.readOnly = true
log.Info("Persisted dirty state to disk", "size", common.StorageSize(journal.Len()), "elapsed", common.PrettyDuration(time.Since(start)))
return nil
}
trie/triedb/pathdb/layertree.go
-214
View File
@@ -1,214 +0,0 @@
// Copyright 2022 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 pathdb
import (
"errors"
"fmt"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// layerTree is a group of state layers identified by the state root.
// This structure defines a few basic operations for manipulating
// state layers linked with each other in a tree structure. It's
// thread-safe to use. However, callers need to ensure the thread-safety
// of the referenced layer by themselves.
type layerTree struct {
lock sync.RWMutex
layers map[common.Hash]layer
}
// newLayerTree constructs the layerTree with the given head layer.
func newLayerTree(head layer) *layerTree {
tree := new(layerTree)
tree.reset(head)
return tree
}
// reset initializes the layerTree by the given head layer.
// All the ancestors will be iterated out and linked in the tree.
func (tree *layerTree) reset(head layer) {
tree.lock.Lock()
defer tree.lock.Unlock()
var layers = make(map[common.Hash]layer)
for head != nil {
layers[head.rootHash()] = head
head = head.parentLayer()
}
tree.layers = layers
}
// get retrieves a layer belonging to the given state root.
func (tree *layerTree) get(root common.Hash) layer {
tree.lock.RLock()
defer tree.lock.RUnlock()
return tree.layers[types.TrieRootHash(root)]
}
// forEach iterates the stored layers inside and applies the
// given callback on them.
func (tree *layerTree) forEach(onLayer func(layer)) {
tree.lock.RLock()
defer tree.lock.RUnlock()
for _, layer := range tree.layers {
onLayer(layer)
}
}
// len returns the number of layers cached.
func (tree *layerTree) len() int {
tree.lock.RLock()
defer tree.lock.RUnlock()
return len(tree.layers)
}
// add inserts a new layer into the tree if it can be linked to an existing old parent.
func (tree *layerTree) add(root common.Hash, parentRoot common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
// Reject noop updates to avoid self-loops. This is a special case that can
// happen for clique networks and proof-of-stake networks where empty blocks
// don't modify the state (0 block subsidy).
//
// Although we could silently ignore this internally, it should be the caller's
// responsibility to avoid even attempting to insert such a layer.
root, parentRoot = types.TrieRootHash(root), types.TrieRootHash(parentRoot)
if root == parentRoot {
return errors.New("layer cycle")
}
parent := tree.get(parentRoot)
if parent == nil {
return fmt.Errorf("triedb parent [%#x] layer missing", parentRoot)
}
l := parent.update(root, parent.stateID()+1, block, nodes.Flatten(), states)
tree.lock.Lock()
tree.layers[l.rootHash()] = l
tree.lock.Unlock()
return nil
}
// cap traverses downwards the diff tree until the number of allowed diff layers
// are crossed. All diffs beyond the permitted number are flattened downwards.
func (tree *layerTree) cap(root common.Hash, layers int) error {
// Retrieve the head layer to cap from
root = types.TrieRootHash(root)
l := tree.get(root)
if l == nil {
return fmt.Errorf("triedb layer [%#x] missing", root)
}
diff, ok := l.(*diffLayer)
if !ok {
return fmt.Errorf("triedb layer [%#x] is disk layer", root)
}
tree.lock.Lock()
defer tree.lock.Unlock()
// If full commit was requested, flatten the diffs and merge onto disk
if layers == 0 {
base, err := diff.persist(true)
if err != nil {
return err
}
// Replace the entire layer tree with the flat base
tree.layers = map[common.Hash]layer{base.rootHash(): base}
return nil
}
// Dive until we run out of layers or reach the persistent database
for i := 0; i < layers-1; i++ {
// If we still have diff layers below, continue down
if parent, ok := diff.parentLayer().(*diffLayer); ok {
diff = parent
} else {
// Diff stack too shallow, return without modifications
return nil
}
}
// 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 := diff.parentLayer().(type) {
case *diskLayer:
return nil
case *diffLayer:
// Hold the lock to prevent any read operations until the new
// parent is linked correctly.
diff.lock.Lock()
base, err := parent.persist(false)
if err != nil {
diff.lock.Unlock()
return err
}
tree.layers[base.rootHash()] = base
diff.parent = base
diff.lock.Unlock()
default:
panic(fmt.Sprintf("unknown data layer in triedb: %T", parent))
}
// Remove any layer that is stale or links into a stale layer
children := make(map[common.Hash][]common.Hash)
for root, layer := range tree.layers {
if dl, ok := layer.(*diffLayer); ok {
parent := dl.parentLayer().rootHash()
children[parent] = append(children[parent], root)
}
}
var remove func(root common.Hash)
remove = func(root common.Hash) {
delete(tree.layers, root)
for _, child := range children[root] {
remove(child)
}
delete(children, root)
}
for root, layer := range tree.layers {
if dl, ok := layer.(*diskLayer); ok && dl.isStale() {
remove(root)
}
}
return nil
}
// bottom returns the bottom-most disk layer in this tree.
func (tree *layerTree) bottom() *diskLayer {
tree.lock.RLock()
defer tree.lock.RUnlock()
if len(tree.layers) == 0 {
return nil // Shouldn't happen, empty tree
}
// pick a random one as the entry point
var current layer
for _, layer := range tree.layers {
current = layer
break
}
for current.parentLayer() != nil {
current = current.parentLayer()
}
return current.(*diskLayer)
}
trie/triedb/pathdb/metrics.go
-50
View File
@@ -1,50 +0,0 @@
// Copyright 2022 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 pathdb
import "github.com/ethereum/go-ethereum/metrics"
var (
cleanHitMeter = metrics.NewRegisteredMeter("pathdb/clean/hit", nil)
cleanMissMeter = metrics.NewRegisteredMeter("pathdb/clean/miss", nil)
cleanReadMeter = metrics.NewRegisteredMeter("pathdb/clean/read", nil)
cleanWriteMeter = metrics.NewRegisteredMeter("pathdb/clean/write", nil)
dirtyHitMeter = metrics.NewRegisteredMeter("pathdb/dirty/hit", nil)
dirtyMissMeter = metrics.NewRegisteredMeter("pathdb/dirty/miss", nil)
dirtyReadMeter = metrics.NewRegisteredMeter("pathdb/dirty/read", nil)
dirtyWriteMeter = metrics.NewRegisteredMeter("pathdb/dirty/write", nil)
dirtyNodeHitDepthHist = metrics.NewRegisteredHistogram("pathdb/dirty/depth", nil, metrics.NewExpDecaySample(1028, 0.015))
cleanFalseMeter = metrics.NewRegisteredMeter("pathdb/clean/false", nil)
dirtyFalseMeter = metrics.NewRegisteredMeter("pathdb/dirty/false", nil)
diskFalseMeter = metrics.NewRegisteredMeter("pathdb/disk/false", nil)
commitTimeTimer = metrics.NewRegisteredTimer("pathdb/commit/time", nil)
commitNodesMeter = metrics.NewRegisteredMeter("pathdb/commit/nodes", nil)
commitBytesMeter = metrics.NewRegisteredMeter("pathdb/commit/bytes", nil)
gcNodesMeter = metrics.NewRegisteredMeter("pathdb/gc/nodes", nil)
gcBytesMeter = metrics.NewRegisteredMeter("pathdb/gc/bytes", nil)
diffLayerBytesMeter = metrics.NewRegisteredMeter("pathdb/diff/bytes", nil)
diffLayerNodesMeter = metrics.NewRegisteredMeter("pathdb/diff/nodes", nil)
historyBuildTimeMeter = metrics.NewRegisteredTimer("pathdb/history/time", nil)
historyDataBytesMeter = metrics.NewRegisteredMeter("pathdb/history/bytes/data", nil)
historyIndexBytesMeter = metrics.NewRegisteredMeter("pathdb/history/bytes/index", nil)
)
trie/triedb/pathdb/nodebuffer.go
-275
View File
@@ -1,275 +0,0 @@
// Copyright 2022 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 pathdb
import (
"fmt"
"time"
"github.com/VictoriaMetrics/fastcache"
"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/trie/trienode"
)
// nodebuffer is a collection of modified trie nodes to aggregate the disk
// write. The content of the nodebuffer must be checked before diving into
// disk (since it basically is not-yet-written data).
type nodebuffer struct {
layers uint64 // The number of diff layers aggregated inside
size uint64 // The size of aggregated writes
limit uint64 // The maximum memory allowance in bytes
nodes map[common.Hash]map[string]*trienode.Node // The dirty node set, mapped by owner and path
}
// newNodeBuffer initializes the node buffer with the provided nodes.
func newNodeBuffer(limit int, nodes map[common.Hash]map[string]*trienode.Node, layers uint64) *nodebuffer {
if nodes == nil {
nodes = make(map[common.Hash]map[string]*trienode.Node)
}
var size uint64
for _, subset := range nodes {
for path, n := range subset {
size += uint64(len(n.Blob) + len(path))
}
}
return &nodebuffer{
layers: layers,
nodes: nodes,
size: size,
limit: uint64(limit),
}
}
// node retrieves the trie node with given node info.
func (b *nodebuffer) node(owner common.Hash, path []byte, hash common.Hash) (*trienode.Node, error) {
subset, ok := b.nodes[owner]
if !ok {
return nil, nil
}
n, ok := subset[string(path)]
if !ok {
return nil, nil
}
if n.Hash != hash {
dirtyFalseMeter.Mark(1)
log.Error("Unexpected trie node in node buffer", "owner", owner, "path", path, "expect", hash, "got", n.Hash)
return nil, newUnexpectedNodeError("dirty", hash, n.Hash, owner, path, n.Blob)
}
return n, nil
}
// commit merges the dirty nodes into the nodebuffer. This operation won't take
// the ownership of the nodes map which belongs to the bottom-most diff layer.
// It will just hold the node references from the given map which are safe to
// copy.
func (b *nodebuffer) commit(nodes map[common.Hash]map[string]*trienode.Node) *nodebuffer {
var (
delta int64
overwrite int64
overwriteSize int64
)
for owner, subset := range nodes {
current, exist := b.nodes[owner]
if !exist {
// Allocate a new map for the subset instead of claiming it directly
// from the passed map to avoid potential concurrent map read/write.
// The nodes belong to original diff layer are still accessible even
// after merging, thus the ownership of nodes map should still belong
// to original layer and any mutation on it should be prevented.
current = make(map[string]*trienode.Node)
for path, n := range subset {
current[path] = n
delta += int64(len(n.Blob) + len(path))
}
b.nodes[owner] = current
continue
}
for path, n := range subset {
if orig, exist := current[path]; !exist {
delta += int64(len(n.Blob) + len(path))
} else {
delta += int64(len(n.Blob) - len(orig.Blob))
overwrite++
overwriteSize += int64(len(orig.Blob) + len(path))
}
current[path] = n
}
b.nodes[owner] = current
}
b.updateSize(delta)
b.layers++
gcNodesMeter.Mark(overwrite)
gcBytesMeter.Mark(overwriteSize)
return b
}
// revert is the reverse operation of commit. It also merges the provided nodes
// into the nodebuffer, the difference is that the provided node set should
// revert the changes made by the last state transition.
func (b *nodebuffer) revert(db ethdb.KeyValueReader, nodes map[common.Hash]map[string]*trienode.Node) error {
// Short circuit if no embedded state transition to revert.
if b.layers == 0 {
return errStateUnrecoverable
}
b.layers--
// Reset the entire buffer if only a single transition left.
if b.layers == 0 {
b.reset()
return nil
}
var delta int64
for owner, subset := range nodes {
current, ok := b.nodes[owner]
if !ok {
panic(fmt.Sprintf("non-existent subset (%x)", owner))
}
for path, n := range subset {
orig, ok := current[path]
if !ok {
// There is a special case in MPT that one child is removed from
// a fullNode which only has two children, and then a new child
// with different position is immediately inserted into the fullNode.
// In this case, the clean child of the fullNode will also be
// marked as dirty because of node collapse and expansion.
//
// In case of database rollback, don't panic if this "clean"
// node occurs which is not present in buffer.
var nhash common.Hash
if owner == (common.Hash{}) {
_, nhash = rawdb.ReadAccountTrieNode(db, []byte(path))
} else {
_, nhash = rawdb.ReadStorageTrieNode(db, owner, []byte(path))
}
// Ignore the clean node in the case described above.
if nhash == n.Hash {
continue
}
panic(fmt.Sprintf("non-existent node (%x %v) blob: %v", owner, path, crypto.Keccak256Hash(n.Blob).Hex()))
}
current[path] = n
delta += int64(len(n.Blob)) - int64(len(orig.Blob))
}
}
b.updateSize(delta)
return nil
}
// updateSize updates the total cache size by the given delta.
func (b *nodebuffer) updateSize(delta int64) {
size := int64(b.size) + delta
if size >= 0 {
b.size = uint64(size)
return
}
s := b.size
b.size = 0
log.Error("Invalid pathdb buffer size", "prev", common.StorageSize(s), "delta", common.StorageSize(delta))
}
// reset cleans up the disk cache.
func (b *nodebuffer) reset() {
b.layers = 0
b.size = 0
b.nodes = make(map[common.Hash]map[string]*trienode.Node)
}
// empty returns an indicator if nodebuffer contains any state transition inside.
func (b *nodebuffer) empty() bool {
return b.layers == 0
}
// setSize sets the buffer size to the provided number, and invokes a flush
// operation if the current memory usage exceeds the new limit.
func (b *nodebuffer) setSize(size int, db ethdb.KeyValueStore, clean *fastcache.Cache, id uint64) error {
b.limit = uint64(size)
return b.flush(db, clean, id, false)
}
// flush persists the in-memory dirty trie node into the disk if the configured
// memory threshold is reached. Note, all data must be written atomically.
func (b *nodebuffer) flush(db ethdb.KeyValueStore, clean *fastcache.Cache, id uint64, force bool) error {
if b.size <= b.limit && !force {
return nil
}
// Ensure the target state id is aligned with the internal counter.
head := rawdb.ReadPersistentStateID(db)
if head+b.layers != id {
return fmt.Errorf("buffer layers (%d) cannot be applied on top of persisted state id (%d) to reach requested state id (%d)", b.layers, head, id)
}
var (
start = time.Now()
batch = db.NewBatchWithSize(int(b.size))
)
nodes := writeNodes(batch, b.nodes, clean)
rawdb.WritePersistentStateID(batch, id)
// Flush all mutations in a single batch
size := batch.ValueSize()
if err := batch.Write(); err != nil {
return err
}
commitBytesMeter.Mark(int64(size))
commitNodesMeter.Mark(int64(nodes))
commitTimeTimer.UpdateSince(start)
log.Debug("Persisted pathdb nodes", "nodes", len(b.nodes), "bytes", common.StorageSize(size), "elapsed", common.PrettyDuration(time.Since(start)))
b.reset()
return nil
}
// writeNodes writes the trie nodes into the provided database batch.
// Note this function will also inject all the newly written nodes
// into clean cache.
func writeNodes(batch ethdb.Batch, nodes map[common.Hash]map[string]*trienode.Node, clean *fastcache.Cache) (total int) {
for owner, subset := range nodes {
for path, n := range subset {
if n.IsDeleted() {
if owner == (common.Hash{}) {
rawdb.DeleteAccountTrieNode(batch, []byte(path))
} else {
rawdb.DeleteStorageTrieNode(batch, owner, []byte(path))
}
if clean != nil {
clean.Del(cacheKey(owner, []byte(path)))
}
} else {
if owner == (common.Hash{}) {
rawdb.WriteAccountTrieNode(batch, []byte(path), n.Blob)
} else {
rawdb.WriteStorageTrieNode(batch, owner, []byte(path), n.Blob)
}
if clean != nil {
clean.Set(cacheKey(owner, []byte(path)), n.Blob)
}
}
}
total += len(subset)
}
return total
}
// cacheKey constructs the unique key of clean cache.
func cacheKey(owner common.Hash, path []byte) []byte {
if owner == (common.Hash{}) {
return path
}
return append(owner.Bytes(), path...)
}
trie/triedb/pathdb/testutils.go
-156
View File
@@ -1,156 +0,0 @@
// Copyright 2023 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 pathdb
import (
"bytes"
"fmt"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
"golang.org/x/exp/slices"
)
// testHasher is a test utility for computing root hash of a batch of state
// elements. The hash algorithm is to sort all the elements in lexicographical
// order, concat the key and value in turn, and perform hash calculation on
// the concatenated bytes. Except the root hash, a nodeset will be returned
// once Commit is called, which contains all the changes made to hasher.
type testHasher struct {
owner common.Hash // owner identifier
root common.Hash // original root
dirties map[common.Hash][]byte // dirty states
cleans map[common.Hash][]byte // clean states
}
// newTestHasher constructs a hasher object with provided states.
func newTestHasher(owner common.Hash, root common.Hash, cleans map[common.Hash][]byte) (*testHasher, error) {
if cleans == nil {
cleans = make(map[common.Hash][]byte)
}
if got, _ := hash(cleans); got != root {
return nil, fmt.Errorf("state root mismatched, want: %x, got: %x", root, got)
}
return &testHasher{
owner: owner,
root: root,
dirties: make(map[common.Hash][]byte),
cleans: cleans,
}, nil
}
// Get returns the value for key stored in the trie.
func (h *testHasher) Get(key []byte) ([]byte, error) {
hash := common.BytesToHash(key)
val, ok := h.dirties[hash]
if ok {
return val, nil
}
return h.cleans[hash], nil
}
// Update associates key with value in the trie.
func (h *testHasher) Update(key, value []byte) error {
h.dirties[common.BytesToHash(key)] = common.CopyBytes(value)
return nil
}
// Delete removes any existing value for key from the trie.
func (h *testHasher) Delete(key []byte) error {
h.dirties[common.BytesToHash(key)] = nil
return nil
}
// Commit computes the new hash of the states and returns the set with all
// state changes.
func (h *testHasher) Commit(collectLeaf bool) (common.Hash, *trienode.NodeSet, error) {
var (
nodes = make(map[common.Hash][]byte)
set = trienode.NewNodeSet(h.owner)
)
for hash, val := range h.cleans {
nodes[hash] = val
}
for hash, val := range h.dirties {
nodes[hash] = val
if bytes.Equal(val, h.cleans[hash]) {
continue
}
if len(val) == 0 {
set.AddNode(hash.Bytes(), trienode.NewDeleted())
} else {
set.AddNode(hash.Bytes(), trienode.New(crypto.Keccak256Hash(val), val))
}
}
root, blob := hash(nodes)
// Include the dirty root node as well.
if root != types.EmptyRootHash && root != h.root {
set.AddNode(nil, trienode.New(root, blob))
}
if root == types.EmptyRootHash && h.root != types.EmptyRootHash {
set.AddNode(nil, trienode.NewDeleted())
}
return root, set, nil
}
// hash performs the hash computation upon the provided states.
func hash(states map[common.Hash][]byte) (common.Hash, []byte) {
var hs []common.Hash
for hash := range states {
hs = append(hs, hash)
}
slices.SortFunc(hs, common.Hash.Cmp)
var input []byte
for _, hash := range hs {
if len(states[hash]) == 0 {
continue
}
input = append(input, hash.Bytes()...)
input = append(input, states[hash]...)
}
if len(input) == 0 {
return types.EmptyRootHash, nil
}
return crypto.Keccak256Hash(input), input
}
type hashLoader struct {
accounts map[common.Hash][]byte
storages map[common.Hash]map[common.Hash][]byte
}
func newHashLoader(accounts map[common.Hash][]byte, storages map[common.Hash]map[common.Hash][]byte) *hashLoader {
return &hashLoader{
accounts: accounts,
storages: storages,
}
}
// OpenTrie opens the main account trie.
func (l *hashLoader) OpenTrie(root common.Hash) (triestate.Trie, error) {
return newTestHasher(common.Hash{}, root, l.accounts)
}
// OpenStorageTrie opens the storage trie of an account.
func (l *hashLoader) OpenStorageTrie(stateRoot common.Hash, addrHash, root common.Hash) (triestate.Trie, error) {
return newTestHasher(addrHash, root, l.storages[addrHash])
}
trie/verkle.go
+2 -4
View File
@@ -26,6 +26,7 @@ import (
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/utils"
"github.com/ethereum/go-ethereum/triedb/database"
"github.com/gballet/go-verkle"
"github.com/holiman/uint256"
)
@@ -39,13 +40,12 @@ var (
// interface so that Verkle trees can be reused verbatim.
type VerkleTrie struct {
root verkle.VerkleNode
db *Database
cache *utils.PointCache
reader *trieReader
}
// NewVerkleTrie constructs a verkle tree based on the specified root hash.
func NewVerkleTrie(root common.Hash, db *Database, cache *utils.PointCache) (*VerkleTrie, error) {
func NewVerkleTrie(root common.Hash, db database.Database, cache *utils.PointCache) (*VerkleTrie, error) {
reader, err := newTrieReader(root, common.Hash{}, db)
if err != nil {
return nil, err
@@ -64,7 +64,6 @@ func NewVerkleTrie(root common.Hash, db *Database, cache *utils.PointCache) (*Ve
}
return &VerkleTrie{
root: node,
db: db,
cache: cache,
reader: reader,
}, nil
@@ -261,7 +260,6 @@ func (t *VerkleTrie) Prove(key []byte, proofDb ethdb.KeyValueWriter) error {
func (t *VerkleTrie) Copy() *VerkleTrie {
return &VerkleTrie{
root: t.root.Copy(),
db: t.db,
cache: t.cache,
reader: t.reader,
}
trie/verkle_test.go
+1 -7
View File
@@ -24,7 +24,6 @@ import (
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/trie/triedb/pathdb"
"github.com/ethereum/go-ethereum/trie/utils"
"github.com/holiman/uint256"
)
@@ -57,12 +56,7 @@ var (
)
func TestVerkleTreeReadWrite(t *testing.T) {
db := NewDatabase(rawdb.NewMemoryDatabase(), &Config{
IsVerkle: true,
PathDB: pathdb.Defaults,
})
defer db.Close()
db := newTestDatabase(rawdb.NewMemoryDatabase(), rawdb.PathScheme)
tr, _ := NewVerkleTrie(types.EmptyVerkleHash, db, utils.NewPointCache(100))
for addr, acct := range accounts {