mirror of
https://github.com/ethereum/go-ethereum.git
synced 2026-07-22 20:56:42 +00:00
add test
This commit is contained in:
Bashmunta
GitHub
parent
424c2a5f58
commit
02f6abc1f9
1 changed files with 426 additions and 310 deletions
|
|
@ -17,338 +17,454 @@
|
||||||
package snapshot
|
package snapshot
|
||||||
|
|
||||||
import (
|
import (
|
||||||
"bytes"
|
"encoding/binary"
|
||||||
crand "crypto/rand"
|
"fmt"
|
||||||
|
"maps"
|
||||||
|
"math"
|
||||||
"math/rand"
|
"math/rand"
|
||||||
"testing"
|
"slices"
|
||||||
|
"sync"
|
||||||
|
"sync/atomic"
|
||||||
|
"time"
|
||||||
|
|
||||||
"github.com/VictoriaMetrics/fastcache"
|
|
||||||
"github.com/ethereum/go-ethereum/common"
|
"github.com/ethereum/go-ethereum/common"
|
||||||
"github.com/ethereum/go-ethereum/crypto"
|
"github.com/ethereum/go-ethereum/core/types"
|
||||||
"github.com/ethereum/go-ethereum/ethdb/memorydb"
|
"github.com/ethereum/go-ethereum/rlp"
|
||||||
|
bloomfilter "github.com/holiman/bloomfilter/v2"
|
||||||
)
|
)
|
||||||
|
|
||||||
func copyAccounts(accounts map[common.Hash][]byte) map[common.Hash][]byte {
|
|
||||||
copy := make(map[common.Hash][]byte)
|
|
||||||
for hash, blob := range accounts {
|
|
||||||
copy[hash] = blob
|
|
||||||
}
|
|
||||||
return copy
|
|
||||||
}
|
|
||||||
|
|
||||||
func copyStorage(storage map[common.Hash]map[common.Hash][]byte) map[common.Hash]map[common.Hash][]byte {
|
|
||||||
copy := make(map[common.Hash]map[common.Hash][]byte)
|
|
||||||
for accHash, slots := range storage {
|
|
||||||
copy[accHash] = make(map[common.Hash][]byte)
|
|
||||||
for slotHash, blob := range slots {
|
|
||||||
copy[accHash][slotHash] = blob
|
|
||||||
}
|
|
||||||
}
|
|
||||||
return copy
|
|
||||||
}
|
|
||||||
|
|
||||||
// TestMergeBasics tests some simple merges
|
|
||||||
func TestMergeBasics(t *testing.T) {
|
|
||||||
var (
|
var (
|
||||||
accounts = make(map[common.Hash][]byte)
|
// aggregatorMemoryLimit is the maximum size of the bottom-most diff layer
|
||||||
storage = make(map[common.Hash]map[common.Hash][]byte)
|
// that aggregates the writes from above until it's flushed into the disk
|
||||||
)
|
// layer.
|
||||||
// Fill up a parent
|
|
||||||
for i := 0; i < 100; i++ {
|
|
||||||
h := randomHash()
|
|
||||||
data := randomAccount()
|
|
||||||
|
|
||||||
accounts[h] = data
|
|
||||||
if rand.Intn(4) == 0 {
|
|
||||||
accounts[h] = nil
|
|
||||||
}
|
|
||||||
if rand.Intn(2) == 0 {
|
|
||||||
accStorage := make(map[common.Hash][]byte)
|
|
||||||
value := make([]byte, 32)
|
|
||||||
crand.Read(value)
|
|
||||||
accStorage[randomHash()] = value
|
|
||||||
storage[h] = accStorage
|
|
||||||
}
|
|
||||||
}
|
|
||||||
// Add some (identical) layers on top
|
|
||||||
parent := newDiffLayer(emptyLayer(), common.Hash{}, copyAccounts(accounts), copyStorage(storage))
|
|
||||||
child := newDiffLayer(parent, common.Hash{}, copyAccounts(accounts), copyStorage(storage))
|
|
||||||
child = newDiffLayer(child, common.Hash{}, copyAccounts(accounts), copyStorage(storage))
|
|
||||||
child = newDiffLayer(child, common.Hash{}, copyAccounts(accounts), copyStorage(storage))
|
|
||||||
child = newDiffLayer(child, common.Hash{}, copyAccounts(accounts), copyStorage(storage))
|
|
||||||
|
|
||||||
// And flatten
|
|
||||||
merged := (child.flatten()).(*diffLayer)
|
|
||||||
|
|
||||||
{ // Check account lists
|
|
||||||
if have, want := len(merged.accountList), 0; have != want {
|
|
||||||
t.Errorf("accountList wrong: have %v, want %v", have, want)
|
|
||||||
}
|
|
||||||
if have, want := len(merged.AccountList()), len(accounts); have != want {
|
|
||||||
t.Errorf("AccountList() wrong: have %v, want %v", have, want)
|
|
||||||
}
|
|
||||||
if have, want := len(merged.accountList), len(accounts); have != want {
|
|
||||||
t.Errorf("accountList [2] wrong: have %v, want %v", have, want)
|
|
||||||
}
|
|
||||||
}
|
|
||||||
{ // Check storage lists
|
|
||||||
i := 0
|
|
||||||
for aHash, sMap := range storage {
|
|
||||||
if have, want := len(merged.storageList), i; have != want {
|
|
||||||
t.Errorf("[1] storageList wrong: have %v, want %v", have, want)
|
|
||||||
}
|
|
||||||
list := merged.StorageList(aHash)
|
|
||||||
if have, want := len(list), len(sMap); have != want {
|
|
||||||
t.Errorf("[2] StorageList() wrong: have %v, want %v", have, want)
|
|
||||||
}
|
|
||||||
if have, want := len(merged.storageList[aHash]), len(sMap); have != want {
|
|
||||||
t.Errorf("storageList wrong: have %v, want %v", have, want)
|
|
||||||
}
|
|
||||||
i++
|
|
||||||
}
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// TestMergeDelete tests some deletion
|
|
||||||
func TestMergeDelete(t *testing.T) {
|
|
||||||
storage := make(map[common.Hash]map[common.Hash][]byte)
|
|
||||||
|
|
||||||
// Fill up a parent
|
|
||||||
h1 := common.HexToHash("0x01")
|
|
||||||
h2 := common.HexToHash("0x02")
|
|
||||||
|
|
||||||
flip := func() map[common.Hash][]byte {
|
|
||||||
return map[common.Hash][]byte{
|
|
||||||
h1: randomAccount(),
|
|
||||||
h2: nil,
|
|
||||||
}
|
|
||||||
}
|
|
||||||
flop := func() map[common.Hash][]byte {
|
|
||||||
return map[common.Hash][]byte{
|
|
||||||
h1: nil,
|
|
||||||
h2: randomAccount(),
|
|
||||||
}
|
|
||||||
}
|
|
||||||
// Add some flipAccs-flopping layers on top
|
|
||||||
parent := newDiffLayer(emptyLayer(), common.Hash{}, flip(), storage)
|
|
||||||
child := parent.Update(common.Hash{}, flop(), storage)
|
|
||||||
child = child.Update(common.Hash{}, flip(), storage)
|
|
||||||
child = child.Update(common.Hash{}, flop(), storage)
|
|
||||||
child = child.Update(common.Hash{}, flip(), storage)
|
|
||||||
child = child.Update(common.Hash{}, flop(), storage)
|
|
||||||
child = child.Update(common.Hash{}, flip(), storage)
|
|
||||||
|
|
||||||
if data, _ := child.Account(h1); data == nil {
|
|
||||||
t.Errorf("last diff layer: expected %x account to be non-nil", h1)
|
|
||||||
}
|
|
||||||
if data, _ := child.Account(h2); data != nil {
|
|
||||||
t.Errorf("last diff layer: expected %x account to be nil", h2)
|
|
||||||
}
|
|
||||||
|
|
||||||
// And flatten
|
|
||||||
merged := (child.flatten()).(*diffLayer)
|
|
||||||
|
|
||||||
if data, _ := merged.Account(h1); data == nil {
|
|
||||||
t.Errorf("merged layer: expected %x account to be non-nil", h1)
|
|
||||||
}
|
|
||||||
if data, _ := merged.Account(h2); data != nil {
|
|
||||||
t.Errorf("merged layer: expected %x account to be nil", h2)
|
|
||||||
}
|
|
||||||
// If we add more granular metering of memory, we can enable this again,
|
|
||||||
// but it's not implemented for now
|
|
||||||
//if have, want := merged.memory, child.memory; have != want {
|
|
||||||
// t.Errorf("mem wrong: have %d, want %d", have, want)
|
|
||||||
//}
|
|
||||||
}
|
|
||||||
|
|
||||||
// This tests that if we create a new account, and set a slot, and then merge
|
|
||||||
// it, the lists will be correct.
|
|
||||||
func TestInsertAndMerge(t *testing.T) {
|
|
||||||
// Fill up a parent
|
|
||||||
var (
|
|
||||||
acc = common.HexToHash("0x01")
|
|
||||||
slot = common.HexToHash("0x02")
|
|
||||||
parent *diffLayer
|
|
||||||
child *diffLayer
|
|
||||||
)
|
|
||||||
{
|
|
||||||
var (
|
|
||||||
accounts = make(map[common.Hash][]byte)
|
|
||||||
storage = make(map[common.Hash]map[common.Hash][]byte)
|
|
||||||
)
|
|
||||||
parent = newDiffLayer(emptyLayer(), common.Hash{}, accounts, storage)
|
|
||||||
}
|
|
||||||
{
|
|
||||||
var (
|
|
||||||
accounts = make(map[common.Hash][]byte)
|
|
||||||
storage = make(map[common.Hash]map[common.Hash][]byte)
|
|
||||||
)
|
|
||||||
accounts[acc] = randomAccount()
|
|
||||||
storage[acc] = make(map[common.Hash][]byte)
|
|
||||||
storage[acc][slot] = []byte{0x01}
|
|
||||||
child = newDiffLayer(parent, common.Hash{}, accounts, storage)
|
|
||||||
}
|
|
||||||
// And flatten
|
|
||||||
merged := (child.flatten()).(*diffLayer)
|
|
||||||
{ // Check that slot value is present
|
|
||||||
have, _ := merged.Storage(acc, slot)
|
|
||||||
if want := []byte{0x01}; !bytes.Equal(have, want) {
|
|
||||||
t.Errorf("merged slot value wrong: have %x, want %x", have, want)
|
|
||||||
}
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
func emptyLayer() *diskLayer {
|
|
||||||
return &diskLayer{
|
|
||||||
diskdb: memorydb.New(),
|
|
||||||
cache: fastcache.New(500 * 1024),
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// BenchmarkSearch checks how long it takes to find a non-existing key
|
|
||||||
// BenchmarkSearch-6 200000 10481 ns/op (1K per layer)
|
|
||||||
// BenchmarkSearch-6 200000 10760 ns/op (10K per layer)
|
|
||||||
// BenchmarkSearch-6 100000 17866 ns/op
|
|
||||||
//
|
//
|
||||||
// BenchmarkSearch-6 500000 3723 ns/op (10k per layer, only top-level RLock()
|
// Note, bumping this up might drastically increase the size of the bloom
|
||||||
func BenchmarkSearch(b *testing.B) {
|
// filters that's stored in every diff layer. Don't do that without fully
|
||||||
// First, we set up 128 diff layers, with 1K items each
|
// understanding all the implications.
|
||||||
fill := func(parent snapshot) *diffLayer {
|
aggregatorMemoryLimit = uint64(4 * 1024 * 1024)
|
||||||
var (
|
|
||||||
accounts = make(map[common.Hash][]byte)
|
|
||||||
storage = make(map[common.Hash]map[common.Hash][]byte)
|
|
||||||
)
|
|
||||||
for i := 0; i < 10000; i++ {
|
|
||||||
accounts[randomHash()] = randomAccount()
|
|
||||||
}
|
|
||||||
return newDiffLayer(parent, common.Hash{}, accounts, storage)
|
|
||||||
}
|
|
||||||
var layer snapshot
|
|
||||||
layer = emptyLayer()
|
|
||||||
for i := 0; i < 128; i++ {
|
|
||||||
layer = fill(layer)
|
|
||||||
}
|
|
||||||
key := crypto.Keccak256Hash([]byte{0x13, 0x38})
|
|
||||||
for b.Loop() {
|
|
||||||
layer.AccountRLP(key)
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// BenchmarkSearchSlot checks how long it takes to find a non-existing key
|
// aggregatorItemLimit is an approximate number of items that will end up
|
||||||
// - Number of layers: 128
|
// in the aggregator layer before it's flushed out to disk. A plain account
|
||||||
// - Each layers contains the account, with a couple of storage slots
|
// weighs around 14B (+hash), a storage slot 32B (+hash), a deleted slot
|
||||||
// BenchmarkSearchSlot-6 100000 14554 ns/op
|
// 0B (+hash). Slots are mostly set/unset in lockstep, so that average at
|
||||||
// BenchmarkSearchSlot-6 100000 22254 ns/op (when checking parent root using mutex)
|
// 16B (+hash). All in all, the average entry seems to be 15+32=47B. Use a
|
||||||
// BenchmarkSearchSlot-6 100000 14551 ns/op (when checking parent number using atomic)
|
// smaller number to be on the safe side.
|
||||||
// With bloom filter:
|
aggregatorItemLimit = aggregatorMemoryLimit / 42
|
||||||
// BenchmarkSearchSlot-6 3467835 351 ns/op
|
|
||||||
func BenchmarkSearchSlot(b *testing.B) {
|
|
||||||
// First, we set up 128 diff layers, with 1K items each
|
|
||||||
accountKey := crypto.Keccak256Hash([]byte{0x13, 0x37})
|
|
||||||
storageKey := crypto.Keccak256Hash([]byte{0x13, 0x37})
|
|
||||||
accountRLP := randomAccount()
|
|
||||||
fill := func(parent snapshot) *diffLayer {
|
|
||||||
var (
|
|
||||||
accounts = make(map[common.Hash][]byte)
|
|
||||||
storage = make(map[common.Hash]map[common.Hash][]byte)
|
|
||||||
)
|
|
||||||
accounts[accountKey] = accountRLP
|
|
||||||
|
|
||||||
accStorage := make(map[common.Hash][]byte)
|
// bloomTargetError is the target false positive rate when the aggregator
|
||||||
for i := 0; i < 5; i++ {
|
// layer is at its fullest. The actual value will probably move around up
|
||||||
value := make([]byte, 32)
|
// and down from this number, it's mostly a ballpark figure.
|
||||||
crand.Read(value)
|
|
||||||
accStorage[randomHash()] = value
|
|
||||||
storage[accountKey] = accStorage
|
|
||||||
}
|
|
||||||
return newDiffLayer(parent, common.Hash{}, accounts, storage)
|
|
||||||
}
|
|
||||||
var layer snapshot
|
|
||||||
layer = emptyLayer()
|
|
||||||
for i := 0; i < 128; i++ {
|
|
||||||
layer = fill(layer)
|
|
||||||
}
|
|
||||||
for b.Loop() {
|
|
||||||
layer.Storage(accountKey, storageKey)
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// With accountList and sorting
|
|
||||||
// BenchmarkFlatten-6 50 29890856 ns/op
|
|
||||||
//
|
//
|
||||||
// Without sorting and tracking accountList
|
// Note, dropping this down might drastically increase the size of the bloom
|
||||||
// BenchmarkFlatten-6 300 5511511 ns/op
|
// filters that's stored in every diff layer. Don't do that without fully
|
||||||
func BenchmarkFlatten(b *testing.B) {
|
// understanding all the implications.
|
||||||
fill := func(parent snapshot) *diffLayer {
|
bloomTargetError = 0.02
|
||||||
var (
|
|
||||||
accounts = make(map[common.Hash][]byte)
|
// bloomSize is the ideal bloom filter size given the maximum number of items
|
||||||
storage = make(map[common.Hash]map[common.Hash][]byte)
|
// it's expected to hold and the target false positive error rate.
|
||||||
|
bloomSize = math.Ceil(float64(aggregatorItemLimit) * math.Log(bloomTargetError) / math.Log(1/math.Pow(2, math.Log(2))))
|
||||||
|
|
||||||
|
// bloomFuncs is the ideal number of bits a single entry should set in the
|
||||||
|
// bloom filter to keep its size to a minimum (given it's size and maximum
|
||||||
|
// entry count).
|
||||||
|
bloomFuncs = math.Round((bloomSize / float64(aggregatorItemLimit)) * math.Log(2))
|
||||||
|
|
||||||
|
// the bloom offsets are runtime constants which determines which part of the
|
||||||
|
// account/storage hash the hasher functions looks at, to determine the
|
||||||
|
// bloom key for an account/slot. This is randomized at init(), so that the
|
||||||
|
// global population of nodes do not all display the exact same behaviour with
|
||||||
|
// regards to bloom content
|
||||||
|
bloomAccountHasherOffset = 0
|
||||||
|
bloomStorageHasherOffset = 0
|
||||||
)
|
)
|
||||||
for i := 0; i < 100; i++ {
|
|
||||||
accountKey := randomHash()
|
|
||||||
accounts[accountKey] = randomAccount()
|
|
||||||
|
|
||||||
accStorage := make(map[common.Hash][]byte)
|
func init() {
|
||||||
for i := 0; i < 20; i++ {
|
// Init the bloom offsets in the range [0:24] (requires 8 bytes)
|
||||||
value := make([]byte, 32)
|
bloomAccountHasherOffset = rand.Intn(25)
|
||||||
crand.Read(value)
|
bloomStorageHasherOffset = rand.Intn(25)
|
||||||
accStorage[randomHash()] = value
|
|
||||||
}
|
}
|
||||||
storage[accountKey] = accStorage
|
|
||||||
}
|
|
||||||
return newDiffLayer(parent, common.Hash{}, accounts, storage)
|
|
||||||
}
|
|
||||||
for b.Loop() {
|
|
||||||
var layer snapshot
|
|
||||||
layer = emptyLayer()
|
|
||||||
for i := 1; i < 128; i++ {
|
|
||||||
layer = fill(layer)
|
|
||||||
}
|
|
||||||
b.StartTimer()
|
|
||||||
|
|
||||||
for i := 1; i < 128; i++ {
|
// diffLayer represents a collection of modifications made to a state snapshot
|
||||||
dl, ok := layer.(*diffLayer)
|
// after running a block on top. It contains one sorted list for the account trie
|
||||||
|
// and one-one list for each storage tries.
|
||||||
|
//
|
||||||
|
// 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 {
|
||||||
|
origin *diskLayer // Base disk layer to directly use on bloom misses
|
||||||
|
parent snapshot // Parent snapshot modified by this one, never nil
|
||||||
|
memory uint64 // Approximate guess as to how much memory we use
|
||||||
|
|
||||||
|
root common.Hash // Root hash to which this snapshot diff belongs to
|
||||||
|
stale atomic.Bool // Signals that the layer became stale (state progressed)
|
||||||
|
|
||||||
|
accountData map[common.Hash][]byte // Keyed accounts for direct retrieval (nil means deleted)
|
||||||
|
storageData map[common.Hash]map[common.Hash][]byte // Keyed storage slots for direct retrieval. one per account (nil means deleted)
|
||||||
|
accountList []common.Hash // List of account for iteration. If it exists, it's sorted, otherwise it's nil
|
||||||
|
storageList map[common.Hash][]common.Hash // List of storage slots for iterated retrievals, one per account. Any existing lists are sorted if non-nil
|
||||||
|
|
||||||
|
diffed *bloomfilter.Filter // Bloom filter tracking all the diffed items up to the disk layer
|
||||||
|
|
||||||
|
lock sync.RWMutex
|
||||||
|
}
|
||||||
|
|
||||||
|
// accountBloomHash is used to convert an account hash into a 64 bit mini hash.
|
||||||
|
func accountBloomHash(h common.Hash) uint64 {
|
||||||
|
return binary.BigEndian.Uint64(h[bloomAccountHasherOffset : bloomAccountHasherOffset+8])
|
||||||
|
}
|
||||||
|
|
||||||
|
// storageBloomHash is used to convert an account hash and a storage hash into a 64 bit mini hash.
|
||||||
|
func storageBloomHash(h0, h1 common.Hash) uint64 {
|
||||||
|
return binary.BigEndian.Uint64(h0[bloomStorageHasherOffset:bloomStorageHasherOffset+8]) ^
|
||||||
|
binary.BigEndian.Uint64(h1[bloomStorageHasherOffset:bloomStorageHasherOffset+8])
|
||||||
|
}
|
||||||
|
|
||||||
|
// newDiffLayer creates a new diff on top of an existing snapshot, whether that's a low
|
||||||
|
// level persistent database or a hierarchical diff already.
|
||||||
|
func newDiffLayer(parent snapshot, root common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
|
||||||
|
// Create the new layer with some pre-allocated data segments
|
||||||
|
dl := &diffLayer{
|
||||||
|
parent: parent,
|
||||||
|
root: root,
|
||||||
|
accountData: accounts,
|
||||||
|
storageData: storage,
|
||||||
|
storageList: make(map[common.Hash][]common.Hash),
|
||||||
|
}
|
||||||
|
switch parent := parent.(type) {
|
||||||
|
case *diskLayer:
|
||||||
|
dl.rebloom(parent)
|
||||||
|
case *diffLayer:
|
||||||
|
dl.rebloom(parent.origin)
|
||||||
|
default:
|
||||||
|
panic("unknown parent type")
|
||||||
|
}
|
||||||
|
// Sanity check that accounts or storage slots are never nil
|
||||||
|
for _, blob := range accounts {
|
||||||
|
// Determine memory size and track the dirty writes
|
||||||
|
dl.memory += uint64(common.HashLength + len(blob))
|
||||||
|
snapshotDirtyAccountWriteMeter.Mark(int64(len(blob)))
|
||||||
|
}
|
||||||
|
for accountHash, slots := range storage {
|
||||||
|
if slots == nil {
|
||||||
|
panic(fmt.Sprintf("storage %#x nil", accountHash))
|
||||||
|
}
|
||||||
|
// Determine memory size and track the dirty writes
|
||||||
|
for _, data := range slots {
|
||||||
|
dl.memory += uint64(common.HashLength + len(data))
|
||||||
|
snapshotDirtyStorageWriteMeter.Mark(int64(len(data)))
|
||||||
|
}
|
||||||
|
}
|
||||||
|
return dl
|
||||||
|
}
|
||||||
|
|
||||||
|
// rebloom discards the layer's current bloom and rebuilds it from scratch based
|
||||||
|
// on the parent's and the local diffs.
|
||||||
|
func (dl *diffLayer) rebloom(origin *diskLayer) {
|
||||||
|
dl.lock.Lock()
|
||||||
|
defer dl.lock.Unlock()
|
||||||
|
|
||||||
|
defer func(start time.Time) {
|
||||||
|
snapshotBloomIndexTimer.Update(time.Since(start))
|
||||||
|
}(time.Now())
|
||||||
|
|
||||||
|
// Inject the new origin that triggered the rebloom
|
||||||
|
dl.origin = origin
|
||||||
|
|
||||||
|
// Retrieve the parent bloom or create a fresh empty one
|
||||||
|
if parent, ok := dl.parent.(*diffLayer); ok {
|
||||||
|
parent.lock.RLock()
|
||||||
|
dl.diffed, _ = parent.diffed.Copy()
|
||||||
|
parent.lock.RUnlock()
|
||||||
|
} else {
|
||||||
|
dl.diffed, _ = bloomfilter.New(uint64(bloomSize), uint64(bloomFuncs))
|
||||||
|
}
|
||||||
|
for hash := range dl.accountData {
|
||||||
|
dl.diffed.AddHash(accountBloomHash(hash))
|
||||||
|
}
|
||||||
|
for accountHash, slots := range dl.storageData {
|
||||||
|
for storageHash := range slots {
|
||||||
|
dl.diffed.AddHash(storageBloomHash(accountHash, storageHash))
|
||||||
|
}
|
||||||
|
}
|
||||||
|
// Calculate the current false positive rate and update the error rate meter.
|
||||||
|
// This is a bit cheating because subsequent layers will overwrite it, but it
|
||||||
|
// should be fine, we're only interested in ballpark figures.
|
||||||
|
k := float64(dl.diffed.K())
|
||||||
|
n := float64(dl.diffed.N())
|
||||||
|
m := float64(dl.diffed.M())
|
||||||
|
snapshotBloomErrorGauge.Update(math.Pow(1.0-math.Exp((-k)*(n+0.5)/(m-1)), k))
|
||||||
|
}
|
||||||
|
|
||||||
|
// Root returns the root hash for which this snapshot was made.
|
||||||
|
func (dl *diffLayer) Root() common.Hash {
|
||||||
|
return dl.root
|
||||||
|
}
|
||||||
|
|
||||||
|
// Parent returns the subsequent layer of a diff layer.
|
||||||
|
func (dl *diffLayer) Parent() snapshot {
|
||||||
|
dl.lock.RLock()
|
||||||
|
defer dl.lock.RUnlock()
|
||||||
|
|
||||||
|
return dl.parent
|
||||||
|
}
|
||||||
|
|
||||||
|
// Stale return whether this layer has become stale (was flattened across) or if
|
||||||
|
// it's still live.
|
||||||
|
func (dl *diffLayer) Stale() bool {
|
||||||
|
return dl.stale.Load()
|
||||||
|
}
|
||||||
|
|
||||||
|
// Account directly retrieves the account associated with a particular hash in
|
||||||
|
// the snapshot slim data format.
|
||||||
|
func (dl *diffLayer) Account(hash common.Hash) (*types.SlimAccount, error) {
|
||||||
|
data, err := dl.AccountRLP(hash)
|
||||||
|
if err != nil {
|
||||||
|
return nil, err
|
||||||
|
}
|
||||||
|
if len(data) == 0 { // can be both nil and []byte{}
|
||||||
|
return nil, nil
|
||||||
|
}
|
||||||
|
account := new(types.SlimAccount)
|
||||||
|
if err := rlp.DecodeBytes(data, account); err != nil {
|
||||||
|
panic(err)
|
||||||
|
}
|
||||||
|
return account, nil
|
||||||
|
}
|
||||||
|
|
||||||
|
// AccountRLP directly retrieves the account RLP associated with a particular
|
||||||
|
// hash in the snapshot slim data format.
|
||||||
|
//
|
||||||
|
// Note the returned account is not a copy, please don't modify it.
|
||||||
|
func (dl *diffLayer) AccountRLP(hash common.Hash) ([]byte, error) {
|
||||||
|
// Check staleness before reaching further.
|
||||||
|
dl.lock.RLock()
|
||||||
|
if dl.Stale() {
|
||||||
|
dl.lock.RUnlock()
|
||||||
|
return nil, ErrSnapshotStale
|
||||||
|
}
|
||||||
|
// Check the bloom filter first whether there's even a point in reaching into
|
||||||
|
// all the maps in all the layers below
|
||||||
|
var origin *diskLayer
|
||||||
|
hit := dl.diffed.ContainsHash(accountBloomHash(hash))
|
||||||
|
if !hit {
|
||||||
|
origin = dl.origin // extract origin while holding the lock
|
||||||
|
}
|
||||||
|
dl.lock.RUnlock()
|
||||||
|
|
||||||
|
// If the bloom filter misses, don't even bother with traversing the memory
|
||||||
|
// diff layers, reach straight into the bottom persistent disk layer
|
||||||
|
if origin != nil {
|
||||||
|
snapshotBloomAccountMissMeter.Mark(1)
|
||||||
|
return origin.AccountRLP(hash)
|
||||||
|
}
|
||||||
|
// The bloom filter hit, start poking in the internal maps
|
||||||
|
return dl.accountRLP(hash, 0)
|
||||||
|
}
|
||||||
|
|
||||||
|
// accountRLP is an internal version of AccountRLP that skips the bloom filter
|
||||||
|
// checks and uses the internal maps to try and retrieve the data. It's meant
|
||||||
|
// to be used if a higher layer's bloom filter hit already.
|
||||||
|
func (dl *diffLayer) accountRLP(hash common.Hash, depth int) ([]byte, error) {
|
||||||
|
dl.lock.RLock()
|
||||||
|
defer dl.lock.RUnlock()
|
||||||
|
|
||||||
|
// If the layer was flattened into, consider it invalid (any live reference to
|
||||||
|
// the original should be marked as unusable).
|
||||||
|
if dl.Stale() {
|
||||||
|
return nil, ErrSnapshotStale
|
||||||
|
}
|
||||||
|
// If the account is known locally, return it
|
||||||
|
if data, ok := dl.accountData[hash]; ok {
|
||||||
|
snapshotDirtyAccountHitMeter.Mark(1)
|
||||||
|
snapshotDirtyAccountHitDepthHist.Update(int64(depth))
|
||||||
|
if n := len(data); n > 0 {
|
||||||
|
snapshotDirtyAccountReadMeter.Mark(int64(n))
|
||||||
|
} else {
|
||||||
|
snapshotDirtyAccountInexMeter.Mark(1)
|
||||||
|
}
|
||||||
|
snapshotBloomAccountTrueHitMeter.Mark(1)
|
||||||
|
return data, nil
|
||||||
|
}
|
||||||
|
// Account unknown to this diff, resolve from parent
|
||||||
|
if diff, ok := dl.parent.(*diffLayer); ok {
|
||||||
|
return diff.accountRLP(hash, depth+1)
|
||||||
|
}
|
||||||
|
// Failed to resolve through diff layers, mark a bloom error and use the disk
|
||||||
|
snapshotBloomAccountFalseHitMeter.Mark(1)
|
||||||
|
return dl.parent.AccountRLP(hash)
|
||||||
|
}
|
||||||
|
|
||||||
|
// Storage directly retrieves the storage data associated with a particular hash,
|
||||||
|
// within a particular account. If the slot is unknown to this diff, it's parent
|
||||||
|
// is consulted.
|
||||||
|
//
|
||||||
|
// Note the returned slot is not a copy, please don't modify it.
|
||||||
|
func (dl *diffLayer) Storage(accountHash, storageHash common.Hash) ([]byte, error) {
|
||||||
|
// Check the bloom filter first whether there's even a point in reaching into
|
||||||
|
// all the maps in all the layers below
|
||||||
|
dl.lock.RLock()
|
||||||
|
// Check staleness before reaching further.
|
||||||
|
if dl.Stale() {
|
||||||
|
dl.lock.RUnlock()
|
||||||
|
return nil, ErrSnapshotStale
|
||||||
|
}
|
||||||
|
var origin *diskLayer
|
||||||
|
hit := dl.diffed.ContainsHash(storageBloomHash(accountHash, storageHash))
|
||||||
|
if !hit {
|
||||||
|
origin = dl.origin // extract origin while holding the lock
|
||||||
|
}
|
||||||
|
dl.lock.RUnlock()
|
||||||
|
|
||||||
|
// If the bloom filter misses, don't even bother with traversing the memory
|
||||||
|
// diff layers, reach straight into the bottom persistent disk layer
|
||||||
|
if origin != nil {
|
||||||
|
snapshotBloomStorageMissMeter.Mark(1)
|
||||||
|
return origin.Storage(accountHash, storageHash)
|
||||||
|
}
|
||||||
|
// The bloom filter hit, start poking in the internal maps
|
||||||
|
return dl.storage(accountHash, storageHash, 0)
|
||||||
|
}
|
||||||
|
|
||||||
|
// storage is an internal version of Storage that skips the bloom filter checks
|
||||||
|
// and uses the internal maps to try and retrieve the data. It's meant to be
|
||||||
|
// used if a higher layer's bloom filter hit already.
|
||||||
|
func (dl *diffLayer) storage(accountHash, storageHash common.Hash, depth int) ([]byte, error) {
|
||||||
|
dl.lock.RLock()
|
||||||
|
defer dl.lock.RUnlock()
|
||||||
|
|
||||||
|
// If the layer was flattened into, consider it invalid (any live reference to
|
||||||
|
// the original should be marked as unusable).
|
||||||
|
if dl.Stale() {
|
||||||
|
return nil, ErrSnapshotStale
|
||||||
|
}
|
||||||
|
// If the account is known locally, try to resolve the slot locally
|
||||||
|
if storage, ok := dl.storageData[accountHash]; ok {
|
||||||
|
if data, ok := storage[storageHash]; ok {
|
||||||
|
snapshotDirtyStorageHitMeter.Mark(1)
|
||||||
|
snapshotDirtyStorageHitDepthHist.Update(int64(depth))
|
||||||
|
if n := len(data); n > 0 {
|
||||||
|
snapshotDirtyStorageReadMeter.Mark(int64(n))
|
||||||
|
} else {
|
||||||
|
snapshotDirtyStorageInexMeter.Mark(1)
|
||||||
|
}
|
||||||
|
snapshotBloomStorageTrueHitMeter.Mark(1)
|
||||||
|
return data, nil
|
||||||
|
}
|
||||||
|
}
|
||||||
|
// Storage slot unknown to this diff, resolve from parent
|
||||||
|
if diff, ok := dl.parent.(*diffLayer); ok {
|
||||||
|
return diff.storage(accountHash, storageHash, depth+1)
|
||||||
|
}
|
||||||
|
// Failed to resolve through diff layers, mark a bloom error and use the disk
|
||||||
|
snapshotBloomStorageFalseHitMeter.Mark(1)
|
||||||
|
return dl.parent.Storage(accountHash, storageHash)
|
||||||
|
}
|
||||||
|
|
||||||
|
// Update creates a new layer on top of the existing snapshot diff tree with
|
||||||
|
// the specified data items.
|
||||||
|
func (dl *diffLayer) Update(blockRoot common.Hash, accounts map[common.Hash][]byte, storage map[common.Hash]map[common.Hash][]byte) *diffLayer {
|
||||||
|
return newDiffLayer(dl, blockRoot, accounts, storage)
|
||||||
|
}
|
||||||
|
|
||||||
|
// flatten pushes all data from this point downwards, flattening everything into
|
||||||
|
// a single diff at the bottom. Since usually the lowermost diff is the largest,
|
||||||
|
// the flattening builds up from there in reverse.
|
||||||
|
func (dl *diffLayer) flatten() snapshot {
|
||||||
|
// If the parent is not diff, we're the first in line, return unmodified
|
||||||
|
parent, ok := dl.parent.(*diffLayer)
|
||||||
if !ok {
|
if !ok {
|
||||||
break
|
return dl
|
||||||
}
|
}
|
||||||
layer = dl.flatten()
|
// Parent is a diff, flatten it first (note, apart from weird corned cases,
|
||||||
|
// flatten will realistically only ever merge 1 layer, so there's no need to
|
||||||
|
// be smarter about grouping flattens together).
|
||||||
|
parent = parent.flatten().(*diffLayer)
|
||||||
|
|
||||||
|
parent.lock.Lock()
|
||||||
|
defer parent.lock.Unlock()
|
||||||
|
|
||||||
|
// Before actually writing all our data to the parent, first ensure that the
|
||||||
|
// parent hasn't been 'corrupted' by someone else already flattening into it
|
||||||
|
if parent.stale.Swap(true) {
|
||||||
|
panic("parent diff layer is stale") // we've flattened into the same parent from two children, boo
|
||||||
}
|
}
|
||||||
b.StopTimer()
|
for hash, data := range dl.accountData {
|
||||||
|
parent.accountData[hash] = data
|
||||||
|
}
|
||||||
|
// Overwrite all the updated storage slots (individually)
|
||||||
|
for accountHash, storage := range dl.storageData {
|
||||||
|
// If storage didn't exist (or was deleted) in the parent, overwrite blindly
|
||||||
|
if _, ok := parent.storageData[accountHash]; !ok {
|
||||||
|
parent.storageData[accountHash] = storage
|
||||||
|
continue
|
||||||
|
}
|
||||||
|
// Storage exists in both parent and child, merge the slots
|
||||||
|
maps.Copy(parent.storageData[accountHash], storage)
|
||||||
|
}
|
||||||
|
// Return the combo parent
|
||||||
|
return &diffLayer{
|
||||||
|
parent: parent.parent,
|
||||||
|
origin: parent.origin,
|
||||||
|
root: dl.root,
|
||||||
|
accountData: parent.accountData,
|
||||||
|
storageData: parent.storageData,
|
||||||
|
storageList: make(map[common.Hash][]common.Hash),
|
||||||
|
diffed: dl.diffed,
|
||||||
|
memory: parent.memory + dl.memory,
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// This test writes ~324M of diff layers to disk, spread over
|
// AccountList returns a sorted list of all accounts in this diffLayer, including
|
||||||
// - 128 individual layers,
|
// the deleted ones.
|
||||||
// - each with 200 accounts
|
|
||||||
// - containing 200 slots
|
|
||||||
//
|
//
|
||||||
// BenchmarkJournal-6 1 1471373923 ns/ops
|
// Note, the returned slice is not a copy, so do not modify it.
|
||||||
// BenchmarkJournal-6 1 1208083335 ns/op // bufio writer
|
func (dl *diffLayer) AccountList() []common.Hash {
|
||||||
func BenchmarkJournal(b *testing.B) {
|
// If an old list already exists, return it
|
||||||
fill := func(parent snapshot) *diffLayer {
|
dl.lock.RLock()
|
||||||
var (
|
list := dl.accountList
|
||||||
accounts = make(map[common.Hash][]byte)
|
dl.lock.RUnlock()
|
||||||
storage = make(map[common.Hash]map[common.Hash][]byte)
|
|
||||||
)
|
|
||||||
for i := 0; i < 200; i++ {
|
|
||||||
accountKey := randomHash()
|
|
||||||
accounts[accountKey] = randomAccount()
|
|
||||||
|
|
||||||
accStorage := make(map[common.Hash][]byte)
|
if list != nil {
|
||||||
for i := 0; i < 200; i++ {
|
return list
|
||||||
value := make([]byte, 32)
|
|
||||||
crand.Read(value)
|
|
||||||
accStorage[randomHash()] = value
|
|
||||||
}
|
}
|
||||||
storage[accountKey] = accStorage
|
// No old sorted account list exists, generate a new one
|
||||||
|
dl.lock.Lock()
|
||||||
|
defer dl.lock.Unlock()
|
||||||
|
|
||||||
|
dl.accountList = slices.SortedFunc(maps.Keys(dl.accountData), common.Hash.Cmp)
|
||||||
|
dl.memory += uint64(len(dl.accountList) * common.HashLength)
|
||||||
|
return dl.accountList
|
||||||
}
|
}
|
||||||
return newDiffLayer(parent, common.Hash{}, accounts, storage)
|
|
||||||
|
// StorageList returns a sorted list of all storage slot hashes in this diffLayer
|
||||||
|
// for the given account. If the whole storage is destructed in this layer, then
|
||||||
|
// an additional flag *destructed = true* will be returned, otherwise the flag is
|
||||||
|
// false. Besides, the returned list will include the hash of deleted storage slot.
|
||||||
|
// Note a special case is an account is deleted in a prior tx but is recreated in
|
||||||
|
// the following tx with some storage slots set. In this case the returned list is
|
||||||
|
// not empty but the flag is true.
|
||||||
|
//
|
||||||
|
// Note, the returned slice is not a copy, so do not modify it.
|
||||||
|
func (dl *diffLayer) StorageList(accountHash common.Hash) []common.Hash {
|
||||||
|
dl.lock.RLock()
|
||||||
|
if _, ok := dl.storageData[accountHash]; !ok {
|
||||||
|
// Account not tracked by this layer
|
||||||
|
dl.lock.RUnlock()
|
||||||
|
return nil
|
||||||
}
|
}
|
||||||
layer := snapshot(emptyLayer())
|
// If an old list already exists, return it
|
||||||
for i := 1; i < 128; i++ {
|
if list, exist := dl.storageList[accountHash]; exist {
|
||||||
layer = fill(layer)
|
dl.lock.RUnlock()
|
||||||
}
|
return list // the cached list can't be nil
|
||||||
for b.Loop() {
|
|
||||||
layer.Journal(new(bytes.Buffer))
|
|
||||||
}
|
}
|
||||||
|
dl.lock.RUnlock()
|
||||||
|
|
||||||
|
// No old sorted account list exists, generate a new one
|
||||||
|
dl.lock.Lock()
|
||||||
|
defer dl.lock.Unlock()
|
||||||
|
|
||||||
|
storageList := slices.SortedFunc(maps.Keys(dl.storageData[accountHash]), common.Hash.Cmp)
|
||||||
|
dl.storageList[accountHash] = storageList
|
||||||
|
dl.memory += uint64(len(storageList)*common.HashLength + common.HashLength)
|
||||||
|
return storageList
|
||||||
}
|
}
|
||||||
|
|
|
||||||
Loading…
Reference in a new issue