go-pulse/trie/iterator.go
Martin Holst Swende 5bf8769fb0
ethdb/memorydb, trie: reduced allocations (#28473)
* trie: use pooling of iterator states in iterator

The node iterator burns through a lot of memory while iterating a trie, and a lot of
that can be avoided by using a fairly small pool (max 40 items).

name        old time/op    new time/op    delta
Iterator-8    6.22ms ± 3%    5.40ms ± 6%  -13.18%  (p=0.008 n=5+5)

name        old alloc/op   new alloc/op   delta
Iterator-8    2.36MB ± 0%    1.67MB ± 0%  -29.23%  (p=0.008 n=5+5)

name        old allocs/op  new allocs/op  delta
Iterator-8     37.0k ± 0%     29.8k ± 0%     ~     (p=0.079 n=4+5)

* ethdb/memorydb: avoid one copying of key

By making the transformation from []byte to string at an earlier point,
we save an allocation which otherwise happens later on.

name           old time/op    new time/op    delta
BatchAllocs-8     412µs ± 6%     382µs ± 2%   -7.18%  (p=0.016 n=5+4)

name           old alloc/op   new alloc/op   delta
BatchAllocs-8     480kB ± 0%     490kB ± 0%   +1.93%  (p=0.008 n=5+5)

name           old allocs/op  new allocs/op  delta
BatchAllocs-8     3.03k ± 0%     2.03k ± 0%  -32.98%  (p=0.008 n=5+5)
2023-11-15 16:20:34 +01:00

792 lines
23 KiB
Go

// Copyright 2014 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 (
"bytes"
"container/heap"
"errors"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
)
// NodeResolver is used for looking up trie nodes before reaching into the real
// persistent layer. This is not mandatory, rather is an optimization for cases
// where trie nodes can be recovered from some external mechanism without reading
// from disk. In those cases, this resolver allows short circuiting accesses and
// returning them from memory.
type NodeResolver func(owner common.Hash, path []byte, hash common.Hash) []byte
// Iterator is a key-value trie iterator that traverses a Trie.
type Iterator struct {
nodeIt NodeIterator
Key []byte // Current data key on which the iterator is positioned on
Value []byte // Current data value on which the iterator is positioned on
Err error
}
// NewIterator creates a new key-value iterator from a node iterator.
// Note that the value returned by the iterator is raw. If the content is encoded
// (e.g. storage value is RLP-encoded), it's caller's duty to decode it.
func NewIterator(it NodeIterator) *Iterator {
return &Iterator{
nodeIt: it,
}
}
// Next moves the iterator forward one key-value entry.
func (it *Iterator) Next() bool {
for it.nodeIt.Next(true) {
if it.nodeIt.Leaf() {
it.Key = it.nodeIt.LeafKey()
it.Value = it.nodeIt.LeafBlob()
return true
}
}
it.Key = nil
it.Value = nil
it.Err = it.nodeIt.Error()
return false
}
// Prove generates the Merkle proof for the leaf node the iterator is currently
// positioned on.
func (it *Iterator) Prove() [][]byte {
return it.nodeIt.LeafProof()
}
// NodeIterator is an iterator to traverse the trie pre-order.
type NodeIterator interface {
// Next moves the iterator to the next node. If the parameter is false, any child
// nodes will be skipped.
Next(bool) bool
// Error returns the error status of the iterator.
Error() error
// Hash returns the hash of the current node.
Hash() common.Hash
// Parent returns the hash of the parent of the current node. The hash may be the one
// grandparent if the immediate parent is an internal node with no hash.
Parent() common.Hash
// Path returns the hex-encoded path to the current node.
// Callers must not retain references to the return value after calling Next.
// For leaf nodes, the last element of the path is the 'terminator symbol' 0x10.
Path() []byte
// NodeBlob returns the rlp-encoded value of the current iterated node.
// If the node is an embedded node in its parent, nil is returned then.
NodeBlob() []byte
// Leaf returns true iff the current node is a leaf node.
Leaf() bool
// LeafKey returns the key of the leaf. The method panics if the iterator is not
// positioned at a leaf. Callers must not retain references to the value after
// calling Next.
LeafKey() []byte
// LeafBlob returns the content of the leaf. The method panics if the iterator
// is not positioned at a leaf. Callers must not retain references to the value
// after calling Next.
LeafBlob() []byte
// LeafProof returns the Merkle proof of the leaf. The method panics if the
// iterator is not positioned at a leaf. Callers must not retain references
// to the value after calling Next.
LeafProof() [][]byte
// AddResolver sets a node resolver to use for looking up trie nodes before
// reaching into the real persistent layer.
//
// This is not required for normal operation, rather is an optimization for
// cases where trie nodes can be recovered from some external mechanism without
// reading from disk. In those cases, this resolver allows short circuiting
// accesses and returning them from memory.
//
// Before adding a similar mechanism to any other place in Geth, consider
// making trie.Database an interface and wrapping at that level. It's a huge
// refactor, but it could be worth it if another occurrence arises.
AddResolver(NodeResolver)
}
// nodeIteratorState represents the iteration state at one particular node of the
// trie, which can be resumed at a later invocation.
type nodeIteratorState struct {
hash common.Hash // Hash of the node being iterated (nil if not standalone)
node node // Trie node being iterated
parent common.Hash // Hash of the first full ancestor node (nil if current is the root)
index int // Child to be processed next
pathlen int // Length of the path to this node
}
type nodeIterator struct {
trie *Trie // Trie being iterated
stack []*nodeIteratorState // Hierarchy of trie nodes persisting the iteration state
path []byte // Path to the current node
err error // Failure set in case of an internal error in the iterator
resolver NodeResolver // optional node resolver for avoiding disk hits
pool []*nodeIteratorState // local pool for iteratorstates
}
// errIteratorEnd is stored in nodeIterator.err when iteration is done.
var errIteratorEnd = errors.New("end of iteration")
// seekError is stored in nodeIterator.err if the initial seek has failed.
type seekError struct {
key []byte
err error
}
func (e seekError) Error() string {
return "seek error: " + e.err.Error()
}
func newNodeIterator(trie *Trie, start []byte) NodeIterator {
if trie.Hash() == types.EmptyRootHash {
return &nodeIterator{
trie: trie,
err: errIteratorEnd,
}
}
it := &nodeIterator{trie: trie}
it.err = it.seek(start)
return it
}
func (it *nodeIterator) putInPool(item *nodeIteratorState) {
if len(it.pool) < 40 {
item.node = nil
it.pool = append(it.pool, item)
}
}
func (it *nodeIterator) getFromPool() *nodeIteratorState {
idx := len(it.pool) - 1
if idx < 0 {
return new(nodeIteratorState)
}
el := it.pool[idx]
it.pool[idx] = nil
it.pool = it.pool[:idx]
return el
}
func (it *nodeIterator) AddResolver(resolver NodeResolver) {
it.resolver = resolver
}
func (it *nodeIterator) Hash() common.Hash {
if len(it.stack) == 0 {
return common.Hash{}
}
return it.stack[len(it.stack)-1].hash
}
func (it *nodeIterator) Parent() common.Hash {
if len(it.stack) == 0 {
return common.Hash{}
}
return it.stack[len(it.stack)-1].parent
}
func (it *nodeIterator) Leaf() bool {
return hasTerm(it.path)
}
func (it *nodeIterator) LeafKey() []byte {
if len(it.stack) > 0 {
if _, ok := it.stack[len(it.stack)-1].node.(valueNode); ok {
return hexToKeybytes(it.path)
}
}
panic("not at leaf")
}
func (it *nodeIterator) LeafBlob() []byte {
if len(it.stack) > 0 {
if node, ok := it.stack[len(it.stack)-1].node.(valueNode); ok {
return node
}
}
panic("not at leaf")
}
func (it *nodeIterator) LeafProof() [][]byte {
if len(it.stack) > 0 {
if _, ok := it.stack[len(it.stack)-1].node.(valueNode); ok {
hasher := newHasher(false)
defer returnHasherToPool(hasher)
proofs := make([][]byte, 0, len(it.stack))
for i, item := range it.stack[:len(it.stack)-1] {
// Gather nodes that end up as hash nodes (or the root)
node, hashed := hasher.proofHash(item.node)
if _, ok := hashed.(hashNode); ok || i == 0 {
proofs = append(proofs, nodeToBytes(node))
}
}
return proofs
}
}
panic("not at leaf")
}
func (it *nodeIterator) Path() []byte {
return it.path
}
func (it *nodeIterator) NodeBlob() []byte {
if it.Hash() == (common.Hash{}) {
return nil // skip the non-standalone node
}
blob, err := it.resolveBlob(it.Hash().Bytes(), it.Path())
if err != nil {
it.err = err
return nil
}
return blob
}
func (it *nodeIterator) Error() error {
if it.err == errIteratorEnd {
return nil
}
if seek, ok := it.err.(seekError); ok {
return seek.err
}
return it.err
}
// Next moves the iterator to the next node, returning whether there are any
// further nodes. In case of an internal error this method returns false and
// sets the Error field to the encountered failure. If `descend` is false,
// skips iterating over any subnodes of the current node.
func (it *nodeIterator) Next(descend bool) bool {
if it.err == errIteratorEnd {
return false
}
if seek, ok := it.err.(seekError); ok {
if it.err = it.seek(seek.key); it.err != nil {
return false
}
}
// Otherwise step forward with the iterator and report any errors.
state, parentIndex, path, err := it.peek(descend)
it.err = err
if it.err != nil {
return false
}
it.push(state, parentIndex, path)
return true
}
func (it *nodeIterator) seek(prefix []byte) error {
// The path we're looking for is the hex encoded key without terminator.
key := keybytesToHex(prefix)
key = key[:len(key)-1]
// Move forward until we're just before the closest match to key.
for {
state, parentIndex, path, err := it.peekSeek(key)
if err == errIteratorEnd {
return errIteratorEnd
} else if err != nil {
return seekError{prefix, err}
} else if bytes.Compare(path, key) >= 0 {
return nil
}
it.push(state, parentIndex, path)
}
}
// init initializes the iterator.
func (it *nodeIterator) init() (*nodeIteratorState, error) {
root := it.trie.Hash()
state := &nodeIteratorState{node: it.trie.root, index: -1}
if root != types.EmptyRootHash {
state.hash = root
}
return state, state.resolve(it, nil)
}
// peek creates the next state of the iterator.
func (it *nodeIterator) peek(descend bool) (*nodeIteratorState, *int, []byte, error) {
// Initialize the iterator if we've just started.
if len(it.stack) == 0 {
state, err := it.init()
return state, nil, nil, err
}
if !descend {
// If we're skipping children, pop the current node first
it.pop()
}
// Continue iteration to the next child
for len(it.stack) > 0 {
parent := it.stack[len(it.stack)-1]
ancestor := parent.hash
if (ancestor == common.Hash{}) {
ancestor = parent.parent
}
state, path, ok := it.nextChild(parent, ancestor)
if ok {
if err := state.resolve(it, path); err != nil {
return parent, &parent.index, path, err
}
return state, &parent.index, path, nil
}
// No more child nodes, move back up.
it.pop()
}
return nil, nil, nil, errIteratorEnd
}
// peekSeek is like peek, but it also tries to skip resolving hashes by skipping
// over the siblings that do not lead towards the desired seek position.
func (it *nodeIterator) peekSeek(seekKey []byte) (*nodeIteratorState, *int, []byte, error) {
// Initialize the iterator if we've just started.
if len(it.stack) == 0 {
state, err := it.init()
return state, nil, nil, err
}
if !bytes.HasPrefix(seekKey, it.path) {
// If we're skipping children, pop the current node first
it.pop()
}
// Continue iteration to the next child
for len(it.stack) > 0 {
parent := it.stack[len(it.stack)-1]
ancestor := parent.hash
if (ancestor == common.Hash{}) {
ancestor = parent.parent
}
state, path, ok := it.nextChildAt(parent, ancestor, seekKey)
if ok {
if err := state.resolve(it, path); err != nil {
return parent, &parent.index, path, err
}
return state, &parent.index, path, nil
}
// No more child nodes, move back up.
it.pop()
}
return nil, nil, nil, errIteratorEnd
}
func (it *nodeIterator) resolveHash(hash hashNode, path []byte) (node, error) {
if it.resolver != nil {
if blob := it.resolver(it.trie.owner, path, common.BytesToHash(hash)); len(blob) > 0 {
if resolved, err := decodeNode(hash, blob); err == nil {
return resolved, nil
}
}
}
// Retrieve the specified node from the underlying node reader.
// it.trie.resolveAndTrack is not used since in that function the
// loaded blob will be tracked, while it's not required here since
// all loaded nodes won't be linked to trie at all and track nodes
// may lead to out-of-memory issue.
blob, err := it.trie.reader.node(path, common.BytesToHash(hash))
if err != nil {
return nil, err
}
// The raw-blob format nodes are loaded either from the
// clean cache or the database, they are all in their own
// copy and safe to use unsafe decoder.
return mustDecodeNodeUnsafe(hash, blob), nil
}
func (it *nodeIterator) resolveBlob(hash hashNode, path []byte) ([]byte, error) {
if it.resolver != nil {
if blob := it.resolver(it.trie.owner, path, common.BytesToHash(hash)); len(blob) > 0 {
return blob, nil
}
}
// Retrieve the specified node from the underlying node reader.
// it.trie.resolveAndTrack is not used since in that function the
// loaded blob will be tracked, while it's not required here since
// all loaded nodes won't be linked to trie at all and track nodes
// may lead to out-of-memory issue.
return it.trie.reader.node(path, common.BytesToHash(hash))
}
func (st *nodeIteratorState) resolve(it *nodeIterator, path []byte) error {
if hash, ok := st.node.(hashNode); ok {
resolved, err := it.resolveHash(hash, path)
if err != nil {
return err
}
st.node = resolved
st.hash = common.BytesToHash(hash)
}
return nil
}
func (it *nodeIterator) findChild(n *fullNode, index int, ancestor common.Hash) (node, *nodeIteratorState, []byte, int) {
var (
path = it.path
child node
state *nodeIteratorState
childPath []byte
)
for ; index < len(n.Children); index++ {
if n.Children[index] != nil {
child = n.Children[index]
hash, _ := child.cache()
state = it.getFromPool()
state.hash = common.BytesToHash(hash)
state.node = child
state.parent = ancestor
state.index = -1
state.pathlen = len(path)
childPath = append(childPath, path...)
childPath = append(childPath, byte(index))
return child, state, childPath, index
}
}
return nil, nil, nil, 0
}
func (it *nodeIterator) nextChild(parent *nodeIteratorState, ancestor common.Hash) (*nodeIteratorState, []byte, bool) {
switch node := parent.node.(type) {
case *fullNode:
// Full node, move to the first non-nil child.
if child, state, path, index := it.findChild(node, parent.index+1, ancestor); child != nil {
parent.index = index - 1
return state, path, true
}
case *shortNode:
// Short node, return the pointer singleton child
if parent.index < 0 {
hash, _ := node.Val.cache()
state := it.getFromPool()
state.hash = common.BytesToHash(hash)
state.node = node.Val
state.parent = ancestor
state.index = -1
state.pathlen = len(it.path)
path := append(it.path, node.Key...)
return state, path, true
}
}
return parent, it.path, false
}
// nextChildAt is similar to nextChild, except that it targets a child as close to the
// target key as possible, thus skipping siblings.
func (it *nodeIterator) nextChildAt(parent *nodeIteratorState, ancestor common.Hash, key []byte) (*nodeIteratorState, []byte, bool) {
switch n := parent.node.(type) {
case *fullNode:
// Full node, move to the first non-nil child before the desired key position
child, state, path, index := it.findChild(n, parent.index+1, ancestor)
if child == nil {
// No more children in this fullnode
return parent, it.path, false
}
// If the child we found is already past the seek position, just return it.
if bytes.Compare(path, key) >= 0 {
parent.index = index - 1
return state, path, true
}
// The child is before the seek position. Try advancing
for {
nextChild, nextState, nextPath, nextIndex := it.findChild(n, index+1, ancestor)
// If we run out of children, or skipped past the target, return the
// previous one
if nextChild == nil || bytes.Compare(nextPath, key) >= 0 {
parent.index = index - 1
return state, path, true
}
// We found a better child closer to the target
state, path, index = nextState, nextPath, nextIndex
}
case *shortNode:
// Short node, return the pointer singleton child
if parent.index < 0 {
hash, _ := n.Val.cache()
state := it.getFromPool()
state.hash = common.BytesToHash(hash)
state.node = n.Val
state.parent = ancestor
state.index = -1
state.pathlen = len(it.path)
path := append(it.path, n.Key...)
return state, path, true
}
}
return parent, it.path, false
}
func (it *nodeIterator) push(state *nodeIteratorState, parentIndex *int, path []byte) {
it.path = path
it.stack = append(it.stack, state)
if parentIndex != nil {
*parentIndex++
}
}
func (it *nodeIterator) pop() {
last := it.stack[len(it.stack)-1]
it.path = it.path[:last.pathlen]
it.stack[len(it.stack)-1] = nil
it.stack = it.stack[:len(it.stack)-1]
// last is now unused
it.putInPool(last)
}
func compareNodes(a, b NodeIterator) int {
if cmp := bytes.Compare(a.Path(), b.Path()); cmp != 0 {
return cmp
}
if a.Leaf() && !b.Leaf() {
return -1
} else if b.Leaf() && !a.Leaf() {
return 1
}
if cmp := bytes.Compare(a.Hash().Bytes(), b.Hash().Bytes()); cmp != 0 {
return cmp
}
if a.Leaf() && b.Leaf() {
return bytes.Compare(a.LeafBlob(), b.LeafBlob())
}
return 0
}
type differenceIterator struct {
a, b NodeIterator // Nodes returned are those in b - a.
eof bool // Indicates a has run out of elements
count int // Number of nodes scanned on either trie
}
// NewDifferenceIterator constructs a NodeIterator that iterates over elements in b that
// are not in a. Returns the iterator, and a pointer to an integer recording the number
// of nodes seen.
func NewDifferenceIterator(a, b NodeIterator) (NodeIterator, *int) {
a.Next(true)
it := &differenceIterator{
a: a,
b: b,
}
return it, &it.count
}
func (it *differenceIterator) Hash() common.Hash {
return it.b.Hash()
}
func (it *differenceIterator) Parent() common.Hash {
return it.b.Parent()
}
func (it *differenceIterator) Leaf() bool {
return it.b.Leaf()
}
func (it *differenceIterator) LeafKey() []byte {
return it.b.LeafKey()
}
func (it *differenceIterator) LeafBlob() []byte {
return it.b.LeafBlob()
}
func (it *differenceIterator) LeafProof() [][]byte {
return it.b.LeafProof()
}
func (it *differenceIterator) Path() []byte {
return it.b.Path()
}
func (it *differenceIterator) NodeBlob() []byte {
return it.b.NodeBlob()
}
func (it *differenceIterator) AddResolver(resolver NodeResolver) {
panic("not implemented")
}
func (it *differenceIterator) Next(bool) bool {
// Invariants:
// - We always advance at least one element in b.
// - At the start of this function, a's path is lexically greater than b's.
if !it.b.Next(true) {
return false
}
it.count++
if it.eof {
// a has reached eof, so we just return all elements from b
return true
}
for {
switch compareNodes(it.a, it.b) {
case -1:
// b jumped past a; advance a
if !it.a.Next(true) {
it.eof = true
return true
}
it.count++
case 1:
// b is before a
return true
case 0:
// a and b are identical; skip this whole subtree if the nodes have hashes
hasHash := it.a.Hash() == common.Hash{}
if !it.b.Next(hasHash) {
return false
}
it.count++
if !it.a.Next(hasHash) {
it.eof = true
return true
}
it.count++
}
}
}
func (it *differenceIterator) Error() error {
if err := it.a.Error(); err != nil {
return err
}
return it.b.Error()
}
type nodeIteratorHeap []NodeIterator
func (h nodeIteratorHeap) Len() int { return len(h) }
func (h nodeIteratorHeap) Less(i, j int) bool { return compareNodes(h[i], h[j]) < 0 }
func (h nodeIteratorHeap) Swap(i, j int) { h[i], h[j] = h[j], h[i] }
func (h *nodeIteratorHeap) Push(x interface{}) { *h = append(*h, x.(NodeIterator)) }
func (h *nodeIteratorHeap) Pop() interface{} {
n := len(*h)
x := (*h)[n-1]
*h = (*h)[0 : n-1]
return x
}
type unionIterator struct {
items *nodeIteratorHeap // Nodes returned are the union of the ones in these iterators
count int // Number of nodes scanned across all tries
}
// NewUnionIterator constructs a NodeIterator that iterates over elements in the union
// of the provided NodeIterators. Returns the iterator, and a pointer to an integer
// recording the number of nodes visited.
func NewUnionIterator(iters []NodeIterator) (NodeIterator, *int) {
h := make(nodeIteratorHeap, len(iters))
copy(h, iters)
heap.Init(&h)
ui := &unionIterator{items: &h}
return ui, &ui.count
}
func (it *unionIterator) Hash() common.Hash {
return (*it.items)[0].Hash()
}
func (it *unionIterator) Parent() common.Hash {
return (*it.items)[0].Parent()
}
func (it *unionIterator) Leaf() bool {
return (*it.items)[0].Leaf()
}
func (it *unionIterator) LeafKey() []byte {
return (*it.items)[0].LeafKey()
}
func (it *unionIterator) LeafBlob() []byte {
return (*it.items)[0].LeafBlob()
}
func (it *unionIterator) LeafProof() [][]byte {
return (*it.items)[0].LeafProof()
}
func (it *unionIterator) Path() []byte {
return (*it.items)[0].Path()
}
func (it *unionIterator) NodeBlob() []byte {
return (*it.items)[0].NodeBlob()
}
func (it *unionIterator) AddResolver(resolver NodeResolver) {
panic("not implemented")
}
// Next returns the next node in the union of tries being iterated over.
//
// It does this by maintaining a heap of iterators, sorted by the iteration
// order of their next elements, with one entry for each source trie. Each
// time Next() is called, it takes the least element from the heap to return,
// advancing any other iterators that also point to that same element. These
// iterators are called with descend=false, since we know that any nodes under
// these nodes will also be duplicates, found in the currently selected iterator.
// Whenever an iterator is advanced, it is pushed back into the heap if it still
// has elements remaining.
//
// In the case that descend=false - eg, we're asked to ignore all subnodes of the
// current node - we also advance any iterators in the heap that have the current
// path as a prefix.
func (it *unionIterator) Next(descend bool) bool {
if len(*it.items) == 0 {
return false
}
// Get the next key from the union
least := heap.Pop(it.items).(NodeIterator)
// Skip over other nodes as long as they're identical, or, if we're not descending, as
// long as they have the same prefix as the current node.
for len(*it.items) > 0 && ((!descend && bytes.HasPrefix((*it.items)[0].Path(), least.Path())) || compareNodes(least, (*it.items)[0]) == 0) {
skipped := heap.Pop(it.items).(NodeIterator)
// Skip the whole subtree if the nodes have hashes; otherwise just skip this node
if skipped.Next(skipped.Hash() == common.Hash{}) {
it.count++
// If there are more elements, push the iterator back on the heap
heap.Push(it.items, skipped)
}
}
if least.Next(descend) {
it.count++
heap.Push(it.items, least)
}
return len(*it.items) > 0
}
func (it *unionIterator) Error() error {
for i := 0; i < len(*it.items); i++ {
if err := (*it.items)[i].Error(); err != nil {
return err
}
}
return nil
}