core, triedb/pathdb: pbss fix release v1.13.8 (continue) and v1.14.0 (#…

…626)

* core/rawdb: improve state scheme checking (#28724)

This pull request improves the condition to check if path state scheme is in use. 

Originally, root node presence was used as the indicator if path scheme is used or not. However due to fact that root node will be deleted during the initial snap sync, this condition is no longer useful.

If PersistentStateID is present, it shows that we've already configured for path scheme.

* core, triedb/pathdb: calculate the size for batch pre-allocation (#29106)

* core, triedb/pathdb: calculate the size for batch pre-allocation

* triedb/pathdb: address comment

* triedb/pathdb: fix panic in recoverable (#29107)

* triedb/pathdb: fix panic in recoverable

* triedb/pathdb: add todo

* triedb/pathdb: rename

* triedb/pathdb: rename

---------

Co-authored-by: rjl493456442 <garyrong0905@gmail.com>

core/rawdb/accessors_trie.go

@@ -294,6 +294,11 @@ func ReadStateScheme(db ethdb.Reader) string {
 	if len(blob) != 0 {
 		return PathScheme
 	}
+	// The root node might be deleted during the initial snap sync, check
+	// the persistent state id then.
+	if id := ReadPersistentStateID(db); id != 0 {
+		return PathScheme
+	}
 	// In a hash-based scheme, the genesis state is consistently stored
 	// on the disk. To assess the scheme of the persistent state, it
 	// suffices to inspect the scheme of the genesis state.

core/rawdb/schema.go

@@ -114,8 +114,8 @@ var (
 	dirtyAccountsKey  = []byte("dacc") // dirtyAccountsPrefix + block hash -> dirty accounts
 
 	// Path-based storage scheme of merkle patricia trie.
-	trieNodeAccountPrefix = []byte("A") // trieNodeAccountPrefix + hexPath -> trie node
-	trieNodeStoragePrefix = []byte("O") // trieNodeStoragePrefix + accountHash + hexPath -> trie node
+	TrieNodeAccountPrefix = []byte("A") // TrieNodeAccountPrefix + hexPath -> trie node
+	TrieNodeStoragePrefix = []byte("O") // TrieNodeStoragePrefix + accountHash + hexPath -> trie node
 	stateIDPrefix         = []byte("L") // stateIDPrefix + state root -> state id
 
 	PreimagePrefix = []byte("secure-key-")       // PreimagePrefix + hash -> preimage
@@ -250,12 +250,12 @@ func genesisStateSpecKey(hash common.Hash) []byte {
 
 // accountTrieNodeKey = trieNodeAccountPrefix + nodePath.
 func accountTrieNodeKey(path []byte) []byte {
-	return append(trieNodeAccountPrefix, path...)
+	return append(TrieNodeAccountPrefix, path...)
 }
 
-// storageTrieNodeKey = trieNodeStoragePrefix + accountHash + nodePath.
+// storageTrieNodeKey = TrieNodeStoragePrefix + accountHash + nodePath.
 func storageTrieNodeKey(accountHash common.Hash, path []byte) []byte {
-	return append(append(trieNodeStoragePrefix, accountHash.Bytes()...), path...)
+	return append(append(TrieNodeStoragePrefix, accountHash.Bytes()...), path...)
 }
 
 func snapshotConsortiumKey(hash common.Hash) []byte {
@@ -277,16 +277,16 @@ func IsLegacyTrieNode(key []byte, val []byte) bool {
 // account trie node in path-based state scheme, and returns the resolved
 // node path if so.
 func ResolveAccountTrieNodeKey(key []byte) (bool, []byte) {
-	if !bytes.HasPrefix(key, trieNodeAccountPrefix) {
+	if !bytes.HasPrefix(key, TrieNodeAccountPrefix) {
 		return false, nil
 	}
 	// The remaining key should only consist a hex node path
 	// whose length is in the range 0 to 64 (64 is excluded
 	// since leaves are always wrapped with shortNode).
-	if len(key) >= len(trieNodeAccountPrefix)+common.HashLength*2 {
+	if len(key) >= len(TrieNodeAccountPrefix)+common.HashLength*2 {
 		return false, nil
 	}
-	return true, key[len(trieNodeAccountPrefix):]
+	return true, key[len(TrieNodeAccountPrefix):]
 }
 
 // IsAccountTrieNode reports whether a provided database entry is an account
@@ -300,20 +300,20 @@ func IsAccountTrieNode(key []byte) bool {
 // trie node in path-based state scheme, and returns the resolved account hash
 // and node path if so.
 func ResolveStorageTrieNode(key []byte) (bool, common.Hash, []byte) {
-	if !bytes.HasPrefix(key, trieNodeStoragePrefix) {
+	if !bytes.HasPrefix(key, TrieNodeStoragePrefix) {
 		return false, common.Hash{}, nil
 	}
 	// The remaining key consists of 2 parts:
 	// - 32 bytes account hash
 	// - hex node path whose length is in the range 0 to 64
-	if len(key) < len(trieNodeStoragePrefix)+common.HashLength {
+	if len(key) < len(TrieNodeStoragePrefix)+common.HashLength {
 		return false, common.Hash{}, nil
 	}
-	if len(key) >= len(trieNodeStoragePrefix)+common.HashLength+common.HashLength*2 {
+	if len(key) >= len(TrieNodeStoragePrefix)+common.HashLength+common.HashLength*2 {
 		return false, common.Hash{}, nil
 	}
-	accountHash := common.BytesToHash(key[len(trieNodeStoragePrefix) : len(trieNodeStoragePrefix)+common.HashLength])
-	return true, accountHash, key[len(trieNodeStoragePrefix)+common.HashLength:]
+	accountHash := common.BytesToHash(key[len(TrieNodeStoragePrefix) : len(TrieNodeStoragePrefix)+common.HashLength])
+	return true, accountHash, key[len(TrieNodeStoragePrefix)+common.HashLength:]
 }
 
 // IsStorageTrieNode reports whether a provided database entry is a storage

core/rawdb/schema_test.go

@@ -99,7 +99,7 @@ func TestResolveAccountTrieNodeKey(t *testing.T) {
 		},
 		{
 			name:          "storage prefixed",
-			inputKey:      append(trieNodeStoragePrefix, bytes4...),
+			inputKey:      append(TrieNodeStoragePrefix, bytes4...),
 			expectedCheck: false,
 			expectedKey:   nil,
 		},
@@ -175,7 +175,7 @@ func TestResolveStorageTrieNode(t *testing.T) {
 		},
 		{
 			name:          "storage prefixed hash 20 length 4",
-			inputKey:      append(append(trieNodeStoragePrefix, bytes20...), bytes4...),
+			inputKey:      append(append(TrieNodeStoragePrefix, bytes20...), bytes4...),
 			expectedCheck: false,
 			expectedHash:  common.Hash{},
 			expectedKey:   nil,

trie/triedb/pathdb/database.go

@@ -380,18 +380,23 @@ func (db *Database) Recoverable(root common.Hash) bool {
 	if *id >= dl.stateID() {
 		return false
 	}
-
+	// This is a temporary workaround for the unavailability of the freezer in
+	// dev mode. As a consequence, the Pathdb loses the ability for deep reorg
+	// in certain cases.
+	// TODO(rjl493456442): Implement the in-memory ancient store.
+	if db.freezer == nil {
+		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 {
+		if m.parent != root {
 			return errors.New("unexpected state history")
 		}
 		if len(m.incomplete) > 0 {
 			return errors.New("incomplete state history")
 		}
-		parent = m.root
+		root = m.root
 		return nil
 	}) == nil
 }

trie/triedb/pathdb/nodebuffer.go

@@ -205,6 +205,19 @@ func (b *nodebuffer) setSize(size int, db ethdb.KeyValueStore, clean *fastcache.
 	return b.flush(db, clean, id, false)
 }
 
+// allocBatch returns a database batch with pre-allocated buffer.
+func (b *nodebuffer) allocBatch(db ethdb.KeyValueStore) ethdb.Batch {
+	var metasize int
+	for owner, nodes := range b.nodes {
+		if owner == (common.Hash{}) {
+			metasize += len(nodes) * len(rawdb.TrieNodeAccountPrefix) // database key prefix
+		} else {
+			metasize += len(nodes) * (len(rawdb.TrieNodeStoragePrefix) + common.HashLength) // database key prefix + owner
+		}
+	}
+	return db.NewBatchWithSize((metasize + int(b.size)) * 11 / 10) // extra 10% for potential pebble internal stuff
+}
+
 // 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 {
@@ -218,7 +231,7 @@ func (b *nodebuffer) flush(db ethdb.KeyValueStore, clean *fastcache.Cache, id ui
 	}
 	var (
 		start = time.Now()
-		batch = db.NewBatchWithSize(int(b.size))
+		batch = b.allocBatch(db)
 	)
 	nodes := writeNodes(batch, b.nodes, clean)
 	rawdb.WritePersistentStateID(batch, id)