From a1b8892384eed99beb3f3364cfd13b049a1c0167 Mon Sep 17 00:00:00 2001 From: rjl493456442 Date: Fri, 19 Aug 2022 06:39:47 +0800 Subject: [PATCH] trie: improve node rlp decoding performance (#25357) This avoids copying the input []byte while decoding trie nodes. In most cases, particularly when the input slice is provided by the underlying database, this optimization is safe to use. For cases where the origin of the input slice is unclear, the copying version is retained. The new code performs better even when the input must be copied, because it is now only copied once in decodeNode. --- trie/database.go | 14 ++++-- trie/node.go | 30 ++++++++++-- trie/node_test.go | 121 ++++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 158 insertions(+), 7 deletions(-) diff --git a/trie/database.go b/trie/database.go index 8c9f47176..8c154ba96 100644 --- a/trie/database.go +++ b/trie/database.go @@ -163,7 +163,10 @@ func (n *cachedNode) rlp() []byte { // or by regenerating it from the rlp encoded blob. func (n *cachedNode) obj(hash common.Hash) node { if node, ok := n.node.(rawNode); ok { - return mustDecodeNode(hash[:], node) + // The raw-blob format nodes are loaded from either from + // clean cache or the database, they are all in their own + // copy and safe to use unsafe decoder. + return mustDecodeNodeUnsafe(hash[:], node) } return expandNode(hash[:], n.node) } @@ -346,7 +349,10 @@ func (db *Database) node(hash common.Hash) node { if enc := db.cleans.Get(nil, hash[:]); enc != nil { memcacheCleanHitMeter.Mark(1) memcacheCleanReadMeter.Mark(int64(len(enc))) - return mustDecodeNode(hash[:], enc) + + // The returned value from cache is in its own copy, + // safe to use mustDecodeNodeUnsafe for decoding. + return mustDecodeNodeUnsafe(hash[:], enc) } } // Retrieve the node from the dirty cache if available @@ -371,7 +377,9 @@ func (db *Database) node(hash common.Hash) node { memcacheCleanMissMeter.Mark(1) memcacheCleanWriteMeter.Mark(int64(len(enc))) } - return mustDecodeNode(hash[:], enc) + // The returned value from database is in its own copy, + // safe to use mustDecodeNodeUnsafe for decoding. + return mustDecodeNodeUnsafe(hash[:], enc) } // Node retrieves an encoded cached trie node from memory. If it cannot be found diff --git a/trie/node.go b/trie/node.go index bf3f024bb..6ce6551de 100644 --- a/trie/node.go +++ b/trie/node.go @@ -99,6 +99,7 @@ func (n valueNode) fstring(ind string) string { return fmt.Sprintf("%x ", []byte(n)) } +// mustDecodeNode is a wrapper of decodeNode and panic if any error is encountered. func mustDecodeNode(hash, buf []byte) node { n, err := decodeNode(hash, buf) if err != nil { @@ -107,8 +108,29 @@ func mustDecodeNode(hash, buf []byte) node { return n } -// decodeNode parses the RLP encoding of a trie node. +// mustDecodeNodeUnsafe is a wrapper of decodeNodeUnsafe and panic if any error is +// encountered. +func mustDecodeNodeUnsafe(hash, buf []byte) node { + n, err := decodeNodeUnsafe(hash, buf) + if err != nil { + panic(fmt.Sprintf("node %x: %v", hash, err)) + } + return n +} + +// decodeNode parses the RLP encoding of a trie node. It will deep-copy the passed +// byte slice for decoding, so it's safe to modify the byte slice afterwards. The- +// decode performance of this function is not optimal, but it is suitable for most +// scenarios with low performance requirements and hard to determine whether the +// byte slice be modified or not. func decodeNode(hash, buf []byte) (node, error) { + return decodeNodeUnsafe(hash, common.CopyBytes(buf)) +} + +// decodeNodeUnsafe parses the RLP encoding of a trie node. The passed byte slice +// will be directly referenced by node without bytes deep copy, so the input MUST +// not be changed after. +func decodeNodeUnsafe(hash, buf []byte) (node, error) { if len(buf) == 0 { return nil, io.ErrUnexpectedEOF } @@ -141,7 +163,7 @@ func decodeShort(hash, elems []byte) (node, error) { if err != nil { return nil, fmt.Errorf("invalid value node: %v", err) } - return &shortNode{key, append(valueNode{}, val...), flag}, nil + return &shortNode{key, valueNode(val), flag}, nil } r, _, err := decodeRef(rest) if err != nil { @@ -164,7 +186,7 @@ func decodeFull(hash, elems []byte) (*fullNode, error) { return n, err } if len(val) > 0 { - n.Children[16] = append(valueNode{}, val...) + n.Children[16] = valueNode(val) } return n, nil } @@ -190,7 +212,7 @@ func decodeRef(buf []byte) (node, []byte, error) { // empty node return nil, rest, nil case kind == rlp.String && len(val) == 32: - return append(hashNode{}, val...), rest, nil + return hashNode(val), rest, nil default: return nil, nil, fmt.Errorf("invalid RLP string size %d (want 0 or 32)", len(val)) } diff --git a/trie/node_test.go b/trie/node_test.go index ac1d8fbef..9b8b33748 100644 --- a/trie/node_test.go +++ b/trie/node_test.go @@ -20,6 +20,7 @@ import ( "bytes" "testing" + "github.com/ethereum/go-ethereum/crypto" "github.com/ethereum/go-ethereum/rlp" ) @@ -92,3 +93,123 @@ func TestDecodeFullNode(t *testing.T) { t.Fatalf("decode full node err: %v", err) } } + +// goos: darwin +// goarch: arm64 +// pkg: github.com/ethereum/go-ethereum/trie +// BenchmarkEncodeShortNode +// BenchmarkEncodeShortNode-8 16878850 70.81 ns/op 48 B/op 1 allocs/op +func BenchmarkEncodeShortNode(b *testing.B) { + node := &shortNode{ + Key: []byte{0x1, 0x2}, + Val: hashNode(randBytes(32)), + } + b.ResetTimer() + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + nodeToBytes(node) + } +} + +// goos: darwin +// goarch: arm64 +// pkg: github.com/ethereum/go-ethereum/trie +// BenchmarkEncodeFullNode +// BenchmarkEncodeFullNode-8 4323273 284.4 ns/op 576 B/op 1 allocs/op +func BenchmarkEncodeFullNode(b *testing.B) { + node := &fullNode{} + for i := 0; i < 16; i++ { + node.Children[i] = hashNode(randBytes(32)) + } + b.ResetTimer() + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + nodeToBytes(node) + } +} + +// goos: darwin +// goarch: arm64 +// pkg: github.com/ethereum/go-ethereum/trie +// BenchmarkDecodeShortNode +// BenchmarkDecodeShortNode-8 7925638 151.0 ns/op 157 B/op 4 allocs/op +func BenchmarkDecodeShortNode(b *testing.B) { + node := &shortNode{ + Key: []byte{0x1, 0x2}, + Val: hashNode(randBytes(32)), + } + blob := nodeToBytes(node) + hash := crypto.Keccak256(blob) + + b.ResetTimer() + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + mustDecodeNode(hash, blob) + } +} + +// goos: darwin +// goarch: arm64 +// pkg: github.com/ethereum/go-ethereum/trie +// BenchmarkDecodeShortNodeUnsafe +// BenchmarkDecodeShortNodeUnsafe-8 9027476 128.6 ns/op 109 B/op 3 allocs/op +func BenchmarkDecodeShortNodeUnsafe(b *testing.B) { + node := &shortNode{ + Key: []byte{0x1, 0x2}, + Val: hashNode(randBytes(32)), + } + blob := nodeToBytes(node) + hash := crypto.Keccak256(blob) + + b.ResetTimer() + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + mustDecodeNodeUnsafe(hash, blob) + } +} + +// goos: darwin +// goarch: arm64 +// pkg: github.com/ethereum/go-ethereum/trie +// BenchmarkDecodeFullNode +// BenchmarkDecodeFullNode-8 1597462 761.9 ns/op 1280 B/op 18 allocs/op +func BenchmarkDecodeFullNode(b *testing.B) { + node := &fullNode{} + for i := 0; i < 16; i++ { + node.Children[i] = hashNode(randBytes(32)) + } + blob := nodeToBytes(node) + hash := crypto.Keccak256(blob) + + b.ResetTimer() + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + mustDecodeNode(hash, blob) + } +} + +// goos: darwin +// goarch: arm64 +// pkg: github.com/ethereum/go-ethereum/trie +// BenchmarkDecodeFullNodeUnsafe +// BenchmarkDecodeFullNodeUnsafe-8 1789070 687.1 ns/op 704 B/op 17 allocs/op +func BenchmarkDecodeFullNodeUnsafe(b *testing.B) { + node := &fullNode{} + for i := 0; i < 16; i++ { + node.Children[i] = hashNode(randBytes(32)) + } + blob := nodeToBytes(node) + hash := crypto.Keccak256(blob) + + b.ResetTimer() + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + mustDecodeNodeUnsafe(hash, blob) + } +}