bsc/trie/sync_bloom.go

190 lines
6.4 KiB
Go
Raw Normal View History

// Copyright 2019 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 (
"encoding/binary"
"fmt"
"sync"
"sync/atomic"
"time"
"github.com/ethereum/go-ethereum/common"
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
"github.com/ethereum/go-ethereum/common/gopool"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/metrics"
bloomfilter "github.com/holiman/bloomfilter/v2"
)
var (
bloomAddMeter = metrics.NewRegisteredMeter("trie/bloom/add", nil)
bloomLoadMeter = metrics.NewRegisteredMeter("trie/bloom/load", nil)
bloomTestMeter = metrics.NewRegisteredMeter("trie/bloom/test", nil)
bloomMissMeter = metrics.NewRegisteredMeter("trie/bloom/miss", nil)
bloomFaultMeter = metrics.NewRegisteredMeter("trie/bloom/fault", nil)
bloomErrorGauge = metrics.NewRegisteredGauge("trie/bloom/error", nil)
)
// SyncBloom is a bloom filter used during fast sync to quickly decide if a trie
// node or contract code already exists on disk or not. It self populates from the
// provided disk database on creation in a background thread and will only start
// returning live results once that's finished.
type SyncBloom struct {
bloom *bloomfilter.Filter
inited uint32
closer sync.Once
closed uint32
pend sync.WaitGroup
}
// NewSyncBloom creates a new bloom filter of the given size (in megabytes) and
// initializes it from the database. The bloom is hard coded to use 3 filters.
func NewSyncBloom(memory uint64, database ethdb.Iteratee) *SyncBloom {
// Create the bloom filter to track known trie nodes
bloom, err := bloomfilter.New(memory*1024*1024*8, 4)
if err != nil {
panic(fmt.Sprintf("failed to create bloom: %v", err))
}
log.Info("Allocated fast sync bloom", "size", common.StorageSize(memory*1024*1024))
// Assemble the fast sync bloom and init it from previous sessions
b := &SyncBloom{
bloom: bloom,
}
b.pend.Add(2)
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
gopool.Submit(func() {
defer b.pend.Done()
b.init(database)
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
})
gopool.Submit(func() {
defer b.pend.Done()
b.meter()
[R4R] performance improvement in many aspects (#257) * focus on performance improvement in many aspects. 1. Do BlockBody verification concurrently; 2. Do calculation of intermediate root concurrently; 3. Preload accounts before processing blocks; 4. Make the snapshot layers configurable. 5. Reuse some object to reduce GC. add * rlp: improve decoder stream implementation (#22858) This commit makes various cleanup changes to rlp.Stream. * rlp: shrink Stream struct This removes a lot of unused padding space in Stream by reordering the fields. The size of Stream changes from 120 bytes to 88 bytes. Stream instances are internally cached and reused using sync.Pool, so this does not improve performance. * rlp: simplify list stack The list stack kept track of the size of the current list context as well as the current offset into it. The size had to be stored in the stack in order to subtract it from the remaining bytes of any enclosing list in ListEnd. It seems that this can be implemented in a simpler way: just subtract the size from the enclosing list context in List instead. * rlp: use atomic.Value for type cache (#22902) All encoding/decoding operations read the type cache to find the writer/decoder function responsible for a type. When analyzing CPU profiles of geth during sync, I found that the use of sync.RWMutex in cache lookups appears in the profiles. It seems we are running into CPU cache contention problems when package rlp is heavily used on all CPU cores during sync. This change makes it use atomic.Value + a writer lock instead of sync.RWMutex. In the common case where the typeinfo entry is present in the cache, we simply fetch the map and lookup the type. * rlp: optimize byte array handling (#22924) This change improves the performance of encoding/decoding [N]byte. name old time/op new time/op delta DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10) EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10) name old alloc/op new alloc/op delta DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10) EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal) * rlp: optimize big.Int decoding for size <= 32 bytes (#22927) This change grows the static integer buffer in Stream to 32 bytes, making it possible to decode 256bit integers without allocating a temporary buffer. In the recent commit 088da24, Stream struct size decreased from 120 bytes down to 88 bytes. This commit grows the struct to 112 bytes again, but the size change will not degrade performance because Stream instances are internally cached in sync.Pool. name old time/op new time/op delta DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10) name old speed new speed delta DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10) * eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841) Getting the raw value is not necessary to decode this type, and decoding it directly from the stream is faster. * fix testcase * debug no lazy * fix can not repair * address comments Co-authored-by: Felix Lange <fjl@twurst.com>
2021-07-29 12:16:53 +03:00
})
return b
}
// init iterates over the database, pushing every trie hash into the bloom filter.
func (b *SyncBloom) init(database ethdb.Iteratee) {
// Iterate over the database, but restart every now and again to avoid holding
// a persistent snapshot since fast sync can push a ton of data concurrently,
// bloating the disk.
//
// Note, this is fine, because everything inserted into leveldb by fast sync is
// also pushed into the bloom directly, so we're not missing anything when the
// iterator is swapped out for a new one.
it := database.NewIterator(nil, nil)
var (
start = time.Now()
swap = time.Now()
)
for it.Next() && atomic.LoadUint32(&b.closed) == 0 {
// If the database entry is a trie node, add it to the bloom
key := it.Key()
if len(key) == common.HashLength {
b.bloom.AddHash(binary.BigEndian.Uint64(key))
bloomLoadMeter.Mark(1)
} else if ok, hash := rawdb.IsCodeKey(key); ok {
// If the database entry is a contract code, add it to the bloom
b.bloom.AddHash(binary.BigEndian.Uint64(hash))
bloomLoadMeter.Mark(1)
}
// If enough time elapsed since the last iterator swap, restart
if time.Since(swap) > 8*time.Second {
key := common.CopyBytes(it.Key())
it.Release()
it = database.NewIterator(nil, key)
log.Info("Initializing state bloom", "items", b.bloom.N(), "errorrate", b.bloom.FalsePosititveProbability(), "elapsed", common.PrettyDuration(time.Since(start)))
swap = time.Now()
}
}
it.Release()
// Mark the bloom filter inited and return
log.Info("Initialized state bloom", "items", b.bloom.N(), "errorrate", b.bloom.FalsePosititveProbability(), "elapsed", common.PrettyDuration(time.Since(start)))
atomic.StoreUint32(&b.inited, 1)
}
// meter periodically recalculates the false positive error rate of the bloom
// filter and reports it in a metric.
func (b *SyncBloom) meter() {
for {
// Report the current error ration. No floats, lame, scale it up.
bloomErrorGauge.Update(int64(b.bloom.FalsePosititveProbability() * 100000))
// Wait one second, but check termination more frequently
for i := 0; i < 10; i++ {
if atomic.LoadUint32(&b.closed) == 1 {
return
}
time.Sleep(100 * time.Millisecond)
}
}
}
// Close terminates any background initializer still running and releases all the
// memory allocated for the bloom.
func (b *SyncBloom) Close() error {
b.closer.Do(func() {
// Ensure the initializer is stopped
atomic.StoreUint32(&b.closed, 1)
b.pend.Wait()
// Wipe the bloom, but mark it "uninited" just in case someone attempts an access
log.Info("Deallocated state bloom", "items", b.bloom.N(), "errorrate", b.bloom.FalsePosititveProbability())
atomic.StoreUint32(&b.inited, 0)
b.bloom = nil
})
return nil
}
// Add inserts a new trie node hash into the bloom filter.
func (b *SyncBloom) Add(hash []byte) {
if atomic.LoadUint32(&b.closed) == 1 {
return
}
b.bloom.AddHash(binary.BigEndian.Uint64(hash))
bloomAddMeter.Mark(1)
}
// Contains tests if the bloom filter contains the given hash:
// - false: the bloom definitely does not contain hash
// - true: the bloom maybe contains hash
//
// While the bloom is being initialized, any query will return true.
func (b *SyncBloom) Contains(hash []byte) bool {
bloomTestMeter.Mark(1)
if atomic.LoadUint32(&b.inited) == 0 {
// We didn't load all the trie nodes from the previous run of Geth yet. As
// such, we can't say for sure if a hash is not present for anything. Until
// the init is done, we're faking "possible presence" for everything.
return true
}
// Bloom initialized, check the real one and report any successful misses
maybe := b.bloom.ContainsHash(binary.BigEndian.Uint64(hash))
if !maybe {
bloomMissMeter.Mark(1)
}
return maybe
}