feat(master): async pruning of orphan nodes

.github/workflows/lint.yml

@@ -15,8 +15,6 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions/setup-go@v5
         with:
-          go-version: '^1.20.0'
+          go-version: 1.21
       - name: golangci-lint
-        uses: golangci/golangci-lint-action@v3
-        with:
-          version: v1.51.2
+        run: make lint

CHANGELOG.md

@@ -1,6 +1,6 @@
 # Changelog
 
-## Unreleased
+## v1.0.0 (October 30, 2023)
 
 ### Improvements
 

batch.go

@@ -1,6 +1,8 @@
 package iavl
 
 import (
+	"sync"
+
 	dbm "github.com/cosmos/cosmos-db"
 )
 
@@ -11,6 +13,7 @@ type BatchWithFlusher struct {
 	db    dbm.DB    // This is only used to create new batch
 	batch dbm.Batch // Batched writing buffer.
 
+	mtx            sync.Mutex
 	flushThreshold int // The threshold to flush the batch to disk.
 }
 
@@ -46,6 +49,9 @@ func (b *BatchWithFlusher) estimateSizeAfterSetting(key []byte, value []byte) (i
 // the batch is flushed to disk, cleared, and a new one is created with buffer pre-allocated to threshold.
 // The addition entry is then added to the batch.
 func (b *BatchWithFlusher) Set(key, value []byte) error {
+	b.mtx.Lock()
+	defer b.mtx.Unlock()
+
 	batchSizeAfter, err := b.estimateSizeAfterSetting(key, value)
 	if err != nil {
 		return err
@@ -67,6 +73,9 @@ func (b *BatchWithFlusher) Set(key, value []byte) error {
 // the batch is flushed to disk, cleared, and a new one is created with buffer pre-allocated to threshold.
 // The deletion entry is then added to the batch.
 func (b *BatchWithFlusher) Delete(key []byte) error {
+	b.mtx.Lock()
+	defer b.mtx.Unlock()
+
 	batchSizeAfter, err := b.estimateSizeAfterSetting(key, []byte{})
 	if err != nil {
 		return err

fastnode/fast_node.go

@@ -30,6 +30,7 @@ func NewNode(key []byte, value []byte, version int64) *Node {
 }
 
 // DeserializeNode constructs an *FastNode from an encoded byte slice.
+// It assumes we do not mutate this input []byte.
 func DeserializeNode(key []byte, buf []byte) (*Node, error) {
 	ver, n, err := encoding.DecodeVarint(buf)
 	if err != nil {

internal/encoding/encoding.go

@@ -30,6 +30,7 @@ var uvarintPool = &sync.Pool{
 
 // decodeBytes decodes a varint length-prefixed byte slice, returning it along with the number
 // of input bytes read.
+// Assumes bz will not be mutated.
 func DecodeBytes(bz []byte) ([]byte, int, error) {
 	s, n, err := DecodeUvarint(bz)
 	if err != nil {
@@ -51,9 +52,9 @@ func DecodeBytes(bz []byte) ([]byte, int, error) {
 	if len(bz) < end {
 		return nil, n, fmt.Errorf("insufficient bytes decoding []byte of length %v", size)
 	}
-	bz2 := make([]byte, size)
-	copy(bz2, bz[n:end])
-	return bz2, end, nil
+	// bz2 := make([]byte, size)
+	// copy(bz2, bz[n:end])
+	return bz[n:end], end, nil
 }
 
 // decodeUvarint decodes a varint-encoded unsigned integer from a byte slice, returning it and the
@@ -97,6 +98,23 @@ func EncodeBytes(w io.Writer, bz []byte) error {
 	return err
 }
 
+var hashLenBz []byte
+
+func init() {
+	hashLenBz = make([]byte, 1)
+	binary.PutUvarint(hashLenBz, 32)
+}
+
+// Encode 32 byte long hash
+func Encode32BytesHash(w io.Writer, bz []byte) error {
+	_, err := w.Write(hashLenBz)
+	if err != nil {
+		return err
+	}
+	_, err = w.Write(bz)
+	return err
+}
+
 // encodeBytesSlice length-prefixes the byte slice and returns it.
 func EncodeBytesSlice(bz []byte) ([]byte, error) {
 	buf := bufPool.Get().(*bytes.Buffer)

keyformat/prefix_formatter.go

@@ -0,0 +1,42 @@
+package keyformat
+
+import "encoding/binary"
+
+// This file builds some dedicated key formatters for what appears in benchmarks.
+
+// Prefixes a single byte before a 32 byte hash.
+type FastPrefixFormatter struct {
+	prefix      byte
+	length      int
+	prefixSlice []byte
+}
+
+func NewFastPrefixFormatter(prefix byte, length int) *FastPrefixFormatter {
+	return &FastPrefixFormatter{prefix: prefix, length: length, prefixSlice: []byte{prefix}}
+}
+
+func (f *FastPrefixFormatter) Key(bz []byte) []byte {
+	key := make([]byte, 1+f.length)
+	key[0] = f.prefix
+	copy(key[1:], bz)
+	return key
+}
+
+func (f *FastPrefixFormatter) Scan(key []byte, a interface{}) {
+	scan(a, key[1:])
+}
+
+func (f *FastPrefixFormatter) KeyInt64(bz int64) []byte {
+	key := make([]byte, 1+f.length)
+	key[0] = f.prefix
+	binary.BigEndian.PutUint64(key[1:], uint64(bz))
+	return key
+}
+
+func (f *FastPrefixFormatter) Prefix() []byte {
+	return f.prefixSlice
+}
+
+func (f *FastPrefixFormatter) Length() int {
+	return 1 + f.length
+}

mutable_tree.go

@@ -266,34 +266,8 @@ func (tree *MutableTree) set(key []byte, value []byte) (updated bool, err error)
 func (tree *MutableTree) recursiveSet(node *Node, key []byte, value []byte) (
 	newSelf *Node, updated bool, err error,
 ) {
-	version := tree.version + 1
-
 	if node.isLeaf() {
-		if !tree.skipFastStorageUpgrade {
-			tree.addUnsavedAddition(key, fastnode.NewNode(key, value, version))
-		}
-		switch bytes.Compare(key, node.key) {
-		case -1: // setKey < leafKey
-			return &Node{
-				key:           node.key,
-				subtreeHeight: 1,
-				size:          2,
-				nodeKey:       nil,
-				leftNode:      NewNode(key, value),
-				rightNode:     node,
-			}, false, nil
-		case 1: // setKey > leafKey
-			return &Node{
-				key:           key,
-				subtreeHeight: 1,
-				size:          2,
-				nodeKey:       nil,
-				leftNode:      node,
-				rightNode:     NewNode(key, value),
-			}, false, nil
-		default:
-			return NewNode(key, value), true, nil
-		}
+		return tree.recursiveSetLeaf(node, key, value)
 	} else {
 		node, err = node.clone(tree)
 		if err != nil {
@@ -327,6 +301,37 @@ func (tree *MutableTree) recursiveSet(node *Node, key []byte, value []byte) (
 	}
 }
 
+func (tree *MutableTree) recursiveSetLeaf(node *Node, key []byte, value []byte) (
+	newSelf *Node, updated bool, err error,
+) {
+	version := tree.version + 1
+	if !tree.skipFastStorageUpgrade {
+		tree.addUnsavedAddition(key, fastnode.NewNode(key, value, version))
+	}
+	switch bytes.Compare(key, node.key) {
+	case -1: // setKey < leafKey
+		return &Node{
+			key:           node.key,
+			subtreeHeight: 1,
+			size:          2,
+			nodeKey:       nil,
+			leftNode:      NewNode(key, value),
+			rightNode:     node,
+		}, false, nil
+	case 1: // setKey > leafKey
+		return &Node{
+			key:           key,
+			subtreeHeight: 1,
+			size:          2,
+			nodeKey:       nil,
+			leftNode:      node,
+			rightNode:     NewNode(key, value),
+		}, false, nil
+	default:
+		return NewNode(key, value), true, nil
+	}
+}
+
 // Remove removes a key from the working tree. The given key byte slice should not be modified
 // after this call, since it may point to data stored inside IAVL.
 func (tree *MutableTree) Remove(key []byte) ([]byte, bool, error) {

node.go

@@ -57,11 +57,15 @@ func GetRootKey(version int64) []byte {
 
 // Node represents a node in a Tree.
 type Node struct {
-	key           []byte
-	value         []byte
-	hash          []byte
-	nodeKey       *NodeKey
-	leftNodeKey   []byte
+	key     []byte
+	value   []byte
+	hash    []byte
+	nodeKey *NodeKey
+	// Legacy: LeftNodeHash
+	// v1: Left node ptr via Version/key
+	leftNodeKey []byte
+	// Legacy: RightNodeHash
+	// v1: Right node ptr via Version/key
 	rightNodeKey  []byte
 	size          int64
 	leftNode      *Node
@@ -517,19 +521,29 @@ func (node *Node) writeHashBytes(w io.Writer, version int64) error {
 		// (e.g. ProofLeafNode.ValueHash)
 		valueHash := sha256.Sum256(node.value)
 
-		err = encoding.EncodeBytes(w, valueHash[:])
+		err = encoding.Encode32BytesHash(w, valueHash[:])
 		if err != nil {
 			return fmt.Errorf("writing value, %w", err)
 		}
 	} else {
-		if node.leftNode == nil || node.rightNode == nil {
+		if (node.leftNode == nil && len(node.leftNodeKey) != 32) || (node.rightNode == nil && len(node.rightNodeKey) != 32) {
 			return ErrEmptyChild
 		}
-		err = encoding.EncodeBytes(w, node.leftNode.hash)
+		// If left/rightNodeKey is 32 bytes, it is a legacy node whose value is just the hash.
+		// We may have skipped fetching leftNode/rightNode.
+		if len(node.leftNodeKey) == 32 {
+			err = encoding.Encode32BytesHash(w, node.leftNodeKey)
+		} else {
+			err = encoding.Encode32BytesHash(w, node.leftNode.hash)
+		}
 		if err != nil {
 			return fmt.Errorf("writing left hash, %w", err)
 		}
-		err = encoding.EncodeBytes(w, node.rightNode.hash)
+		if len(node.rightNodeKey) == 32 {
+			err = encoding.Encode32BytesHash(w, node.rightNodeKey)
+		} else {
+			err = encoding.Encode32BytesHash(w, node.rightNode.hash)
+		}
 		if err != nil {
 			return fmt.Errorf("writing right hash, %w", err)
 		}