From 4ea4d2dc3473afd9d2eda6ef6b359accce1f0946 Mon Sep 17 00:00:00 2001 From: Zsolt Felfoldi Date: Fri, 18 Aug 2017 21:52:20 +0200 Subject: [PATCH] core, eth: add bloombit indexer, filter based on it --- core/blockchain.go | 15 - core/bloombits/fetcher_test.go | 101 ++++++ core/bloombits/matcher.go | 579 ++++++++++++++++++++++++++++++ core/bloombits/matcher_test.go | 196 ++++++++++ core/bloombits/utils.go | 63 ++++ core/chain_indexer.go | 76 +++- core/chain_indexer_test.go | 7 +- core/database_util.go | 69 ++-- core/database_util_test.go | 108 ------ core/types/bloom9.go | 14 + eth/api_backend.go | 27 ++ eth/backend.go | 10 +- eth/backend_test.go | 74 ---- eth/db_upgrade.go | 68 ++-- eth/filters/api.go | 48 ++- eth/filters/bench_test.go | 237 ++++++++++++ eth/filters/filter.go | 232 +++++++----- eth/filters/filter_system_test.go | 50 ++- eth/filters/filter_test.go | 70 +--- eth/handler.go | 5 + les/api_backend.go | 18 + les/backend.go | 2 +- miner/worker.go | 2 - 23 files changed, 1589 insertions(+), 482 deletions(-) create mode 100644 core/bloombits/fetcher_test.go create mode 100644 core/bloombits/matcher.go create mode 100644 core/bloombits/matcher_test.go create mode 100644 core/bloombits/utils.go delete mode 100644 eth/backend_test.go create mode 100644 eth/filters/bench_test.go diff --git a/core/blockchain.go b/core/blockchain.go index 0bb12fc19..d74b3520b 100644 --- a/core/blockchain.go +++ b/core/blockchain.go @@ -759,12 +759,6 @@ func (bc *BlockChain) InsertReceiptChain(blockChain types.Blocks, receiptChain [ log.Crit("Failed to write block receipts", "err", err) return } - if err := WriteMipmapBloom(bc.chainDb, block.NumberU64(), receipts); err != nil { - errs[index] = fmt.Errorf("failed to write log blooms: %v", err) - atomic.AddInt32(&failed, 1) - log.Crit("Failed to write log blooms", "err", err) - return - } if err := WriteTxLookupEntries(bc.chainDb, block); err != nil { errs[index] = fmt.Errorf("failed to write lookup metadata: %v", err) atomic.AddInt32(&failed, 1) @@ -1017,10 +1011,6 @@ func (bc *BlockChain) InsertChain(chain types.Blocks) (int, error) { if err := WriteTxLookupEntries(bc.chainDb, block); err != nil { return i, err } - // Write map map bloom filters - if err := WriteMipmapBloom(bc.chainDb, block.NumberU64(), receipts); err != nil { - return i, err - } // Write hash preimages if err := WritePreimages(bc.chainDb, block.NumberU64(), state.Preimages()); err != nil { return i, err @@ -1178,11 +1168,6 @@ func (bc *BlockChain) reorg(oldBlock, newBlock *types.Block) error { if err := WriteTxLookupEntries(bc.chainDb, block); err != nil { return err } - // Write map map bloom filters - receipts := GetBlockReceipts(bc.chainDb, block.Hash(), block.NumberU64()) - if err := WriteMipmapBloom(bc.chainDb, block.NumberU64(), receipts); err != nil { - return err - } addedTxs = append(addedTxs, block.Transactions()...) } diff --git a/core/bloombits/fetcher_test.go b/core/bloombits/fetcher_test.go new file mode 100644 index 000000000..9c229cf8d --- /dev/null +++ b/core/bloombits/fetcher_test.go @@ -0,0 +1,101 @@ +// Copyright 2017 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 . +package bloombits + +import ( + "bytes" + "encoding/binary" + "math/rand" + "sync" + "sync/atomic" + "testing" + "time" +) + +const testFetcherReqCount = 5000 + +func fetcherTestVector(b uint, s uint64) []byte { + r := make([]byte, 10) + binary.BigEndian.PutUint16(r[0:2], uint16(b)) + binary.BigEndian.PutUint64(r[2:10], s) + return r +} + +func TestFetcher(t *testing.T) { + testFetcher(t, 1) +} + +func TestFetcherMultipleReaders(t *testing.T) { + testFetcher(t, 10) +} + +func testFetcher(t *testing.T, cnt int) { + f := &fetcher{ + requestMap: make(map[uint64]fetchRequest), + } + distCh := make(chan distRequest, channelCap) + stop := make(chan struct{}) + var reqCount uint32 + + for i := 0; i < 10; i++ { + go func() { + for { + req, ok := <-distCh + if !ok { + return + } + time.Sleep(time.Duration(rand.Intn(100000))) + atomic.AddUint32(&reqCount, 1) + f.deliver([]uint64{req.sectionIndex}, [][]byte{fetcherTestVector(req.bloomIndex, req.sectionIndex)}) + } + }() + } + + var wg, wg2 sync.WaitGroup + for cc := 0; cc < cnt; cc++ { + wg.Add(1) + in := make(chan uint64, channelCap) + out := f.fetch(in, distCh, stop, &wg2) + + time.Sleep(time.Millisecond * 10 * time.Duration(cc)) + go func() { + for i := uint64(0); i < testFetcherReqCount; i++ { + in <- i + } + }() + + go func() { + for i := uint64(0); i < testFetcherReqCount; i++ { + bv := <-out + if !bytes.Equal(bv, fetcherTestVector(0, i)) { + if len(bv) != 10 { + t.Errorf("Vector #%d length is %d, expected 10", i, len(bv)) + } else { + j := binary.BigEndian.Uint64(bv[2:10]) + t.Errorf("Expected vector #%d, fetched #%d", i, j) + } + } + } + wg.Done() + }() + } + + wg.Wait() + close(stop) + if reqCount != testFetcherReqCount { + t.Errorf("Request count mismatch: expected %v, got %v", testFetcherReqCount, reqCount) + } +} diff --git a/core/bloombits/matcher.go b/core/bloombits/matcher.go new file mode 100644 index 000000000..5a7df6b1c --- /dev/null +++ b/core/bloombits/matcher.go @@ -0,0 +1,579 @@ +// Copyright 2017 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 . +package bloombits + +import ( + "sync" + + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/common/bitutil" + "github.com/ethereum/go-ethereum/core/types" +) + +const channelCap = 100 + +// fetcher handles bit vector retrieval pipelines for a single bit index +type fetcher struct { + bloomIndex uint + requestMap map[uint64]fetchRequest + requestLock sync.RWMutex +} + +// fetchRequest represents the state of a bit vector requested from a fetcher. When a distRequest has been sent to the distributor but +// the data has not been delivered yet, queued is true. When delivered, it is stored in the data field and the delivered channel is closed. +type fetchRequest struct { + data []byte + queued bool + delivered chan struct{} +} + +// distRequest is sent by the fetcher to the distributor which groups and prioritizes these requests. +type distRequest struct { + bloomIndex uint + sectionIndex uint64 +} + +// fetch creates a retrieval pipeline, receiving section indexes from sectionCh and returning the results +// in the same order through the returned channel. Multiple fetch instances of the same fetcher are allowed +// to run in parallel, in case the same bit index appears multiple times in the filter structure. Each section +// is requested only once, requests are sent to the request distributor (part of Matcher) through distCh. +func (f *fetcher) fetch(sectionCh chan uint64, distCh chan distRequest, stop chan struct{}, wg *sync.WaitGroup) chan []byte { + dataCh := make(chan []byte, channelCap) + returnCh := make(chan uint64, channelCap) + wg.Add(2) + + go func() { + defer wg.Done() + defer close(returnCh) + + for { + select { + case <-stop: + return + case idx, ok := <-sectionCh: + if !ok { + return + } + + req := false + f.requestLock.Lock() + r := f.requestMap[idx] + if r.data == nil { + req = !r.queued + r.queued = true + if r.delivered == nil { + r.delivered = make(chan struct{}) + } + f.requestMap[idx] = r + } + f.requestLock.Unlock() + if req { + distCh <- distRequest{bloomIndex: f.bloomIndex, sectionIndex: idx} // success is guaranteed, distibuteRequests shuts down after fetch + } + select { + case <-stop: + return + case returnCh <- idx: + } + } + } + }() + + go func() { + defer wg.Done() + defer close(dataCh) + + for { + select { + case <-stop: + return + case idx, ok := <-returnCh: + if !ok { + return + } + + f.requestLock.RLock() + r := f.requestMap[idx] + f.requestLock.RUnlock() + + if r.data == nil { + select { + case <-stop: + return + case <-r.delivered: + f.requestLock.RLock() + r = f.requestMap[idx] + f.requestLock.RUnlock() + } + } + select { + case <-stop: + return + case dataCh <- r.data: + } + } + } + }() + + return dataCh +} + +// deliver is called by the request distributor when a reply to a request has +// arrived +func (f *fetcher) deliver(sectionIdxList []uint64, data [][]byte) { + f.requestLock.Lock() + defer f.requestLock.Unlock() + + for i, sectionIdx := range sectionIdxList { + r := f.requestMap[sectionIdx] + if r.data != nil { + panic("BloomBits section data delivered twice") + } + r.data = data[i] + close(r.delivered) + f.requestMap[sectionIdx] = r + } +} + +// Matcher is a pipelined structure of fetchers and logic matchers which perform +// binary AND/OR operations on the bitstreams, finally creating a stream of potential matches. +type Matcher struct { + addresses []types.BloomIndexList + topics [][]types.BloomIndexList + fetchers map[uint]*fetcher + sectionSize uint64 + + distCh chan distRequest + reqs map[uint][]uint64 + freeQueues map[uint]struct{} + allocQueue []chan uint + running bool + stop chan struct{} + lock sync.Mutex + wg, distWg sync.WaitGroup +} + +// NewMatcher creates a new Matcher instance +func NewMatcher(sectionSize uint64, addresses []common.Address, topics [][]common.Hash) *Matcher { + m := &Matcher{ + fetchers: make(map[uint]*fetcher), + reqs: make(map[uint][]uint64), + freeQueues: make(map[uint]struct{}), + distCh: make(chan distRequest, channelCap), + sectionSize: sectionSize, + } + m.setAddresses(addresses) + m.setTopics(topics) + return m +} + +// setAddresses matches only logs that are generated from addresses that are included +// in the given addresses. +func (m *Matcher) setAddresses(addresses []common.Address) { + m.addresses = make([]types.BloomIndexList, len(addresses)) + for i, address := range addresses { + m.addresses[i] = types.BloomIndexes(address.Bytes()) + } + + for _, bloomIndexList := range m.addresses { + for _, bloomIndex := range bloomIndexList { + m.newFetcher(bloomIndex) + } + } +} + +// setTopics matches only logs that have topics matching the given topics. +func (m *Matcher) setTopics(topics [][]common.Hash) { + m.topics = nil +loop: + for _, topicList := range topics { + t := make([]types.BloomIndexList, len(topicList)) + for i, topic := range topicList { + if (topic == common.Hash{}) { + continue loop + } + t[i] = types.BloomIndexes(topic.Bytes()) + } + m.topics = append(m.topics, t) + } + + for _, bloomIndexLists := range m.topics { + for _, bloomIndexList := range bloomIndexLists { + for _, bloomIndex := range bloomIndexList { + m.newFetcher(bloomIndex) + } + } + } +} + +// match creates a daisy-chain of sub-matchers, one for the address set and one for each topic set, each +// sub-matcher receiving a section only if the previous ones have all found a potential match in one of +// the blocks of the section, then binary AND-ing its own matches and forwaring the result to the next one +func (m *Matcher) match(processCh chan partialMatches) chan partialMatches { + indexLists := m.topics + if len(m.addresses) > 0 { + indexLists = append([][]types.BloomIndexList{m.addresses}, indexLists...) + } + m.distributeRequests() + + for _, subIndexList := range indexLists { + processCh = m.subMatch(processCh, subIndexList) + } + return processCh +} + +// partialMatches with a non-nil vector represents a section in which some sub-matchers have already +// found potential matches. Subsequent sub-matchers will binary AND their matches with this vector. +// If vector is nil, it represents a section to be processed by the first sub-matcher. +type partialMatches struct { + sectionIndex uint64 + vector []byte +} + +// newFetcher adds a fetcher for the given bit index if it has not existed before +func (m *Matcher) newFetcher(idx uint) { + if _, ok := m.fetchers[idx]; ok { + return + } + f := &fetcher{ + bloomIndex: idx, + requestMap: make(map[uint64]fetchRequest), + } + m.fetchers[idx] = f +} + +// subMatch creates a sub-matcher that filters for a set of addresses or topics, binary OR-s those matches, then +// binary AND-s the result to the daisy-chain input (processCh) and forwards it to the daisy-chain output. +// The matches of each address/topic are calculated by fetching the given sections of the three bloom bit indexes belonging to +// that address/topic, and binary AND-ing those vectors together. +func (m *Matcher) subMatch(processCh chan partialMatches, bloomIndexLists []types.BloomIndexList) chan partialMatches { + // set up fetchers + fetchIndexChannels := make([][3]chan uint64, len(bloomIndexLists)) + fetchDataChannels := make([][3]chan []byte, len(bloomIndexLists)) + for i, bloomIndexList := range bloomIndexLists { + for j, bloomIndex := range bloomIndexList { + fetchIndexChannels[i][j] = make(chan uint64, channelCap) + fetchDataChannels[i][j] = m.fetchers[bloomIndex].fetch(fetchIndexChannels[i][j], m.distCh, m.stop, &m.wg) + } + } + + fetchedCh := make(chan partialMatches, channelCap) // entries from processCh are forwarded here after fetches have been initiated + resultsCh := make(chan partialMatches, channelCap) + + m.wg.Add(2) + // goroutine for starting retrievals + go func() { + defer m.wg.Done() + + for { + select { + case <-m.stop: + return + case s, ok := <-processCh: + if !ok { + close(fetchedCh) + for _, fetchIndexChs := range fetchIndexChannels { + for _, fetchIndexCh := range fetchIndexChs { + close(fetchIndexCh) + } + } + return + } + + for _, fetchIndexChs := range fetchIndexChannels { + for _, fetchIndexCh := range fetchIndexChs { + select { + case <-m.stop: + return + case fetchIndexCh <- s.sectionIndex: + } + } + } + select { + case <-m.stop: + return + case fetchedCh <- s: + } + } + } + }() + + // goroutine for processing retrieved data + go func() { + defer m.wg.Done() + + for { + select { + case <-m.stop: + return + case s, ok := <-fetchedCh: + if !ok { + close(resultsCh) + return + } + + var orVector []byte + for _, fetchDataChs := range fetchDataChannels { + var andVector []byte + for _, fetchDataCh := range fetchDataChs { + var data []byte + select { + case <-m.stop: + return + case data = <-fetchDataCh: + } + if andVector == nil { + andVector = make([]byte, int(m.sectionSize/8)) + copy(andVector, data) + } else { + bitutil.ANDBytes(andVector, andVector, data) + } + } + if orVector == nil { + orVector = andVector + } else { + bitutil.ORBytes(orVector, orVector, andVector) + } + } + + if orVector == nil { + orVector = make([]byte, int(m.sectionSize/8)) + } + if s.vector != nil { + bitutil.ANDBytes(orVector, orVector, s.vector) + } + if bitutil.TestBytes(orVector) { + select { + case <-m.stop: + return + case resultsCh <- partialMatches{s.sectionIndex, orVector}: + } + } + } + } + }() + + return resultsCh +} + +// Start starts the matching process and returns a stream of bloom matches in +// a given range of blocks. +// It returns a results channel immediately and stops if Stop is called or there +// are no more matches in the range (in which case the results channel is closed). +// Start/Stop can be called multiple times for different ranges, in which case already +// delivered bit vectors are not requested again. +func (m *Matcher) Start(begin, end uint64) chan uint64 { + m.stop = make(chan struct{}) + processCh := make(chan partialMatches, channelCap) + resultsCh := make(chan uint64, channelCap) + + res := m.match(processCh) + + startSection := begin / m.sectionSize + endSection := end / m.sectionSize + + m.wg.Add(2) + go func() { + defer m.wg.Done() + defer close(processCh) + + for i := startSection; i <= endSection; i++ { + select { + case processCh <- partialMatches{i, nil}: + case <-m.stop: + return + } + } + }() + + go func() { + defer m.wg.Done() + defer close(resultsCh) + + for { + select { + case r, ok := <-res: + if !ok { + return + } + sectionStart := r.sectionIndex * m.sectionSize + s := sectionStart + if begin > s { + s = begin + } + e := sectionStart + m.sectionSize - 1 + if end < e { + e = end + } + for i := s; i <= e; i++ { + b := r.vector[(i-sectionStart)/8] + bit := 7 - i%8 + if b != 0 { + if b&(1< queue[i] { + i++ + } + queue = append(queue, 0) + copy(queue[i+1:], queue[i:len(queue)-1]) + queue[i] = r.sectionIndex + m.reqs[r.bloomIndex] = queue + if len(queue) == 1 { + m.freeQueue(r.bloomIndex) + } + m.lock.Unlock() + case <-stopDist: + m.lock.Lock() + for _, ch := range m.allocQueue { + close(ch) + } + m.allocQueue = nil + m.running = false + m.lock.Unlock() + m.distWg.Done() + return + } + } + }() +} + +// freeQueue marks a queue as free if there are no AllocSectionQueue functions +// waiting for allocation. If there is someone waiting, the queue is immediately +// allocated. +func (m *Matcher) freeQueue(bloomIndex uint) { + if len(m.allocQueue) > 0 { + m.allocQueue[0] <- bloomIndex + m.allocQueue = m.allocQueue[1:] + } else { + m.freeQueues[bloomIndex] = struct{}{} + } +} + +// AllocSectionQueue allocates a queue of requested section indexes belonging to the same +// bloom bit index for a client process that can either immediately fetch the contents +// of the queue or wait a little while for more section indexes to be requested. +func (m *Matcher) AllocSectionQueue() (uint, bool) { + m.lock.Lock() + if !m.running { + m.lock.Unlock() + return 0, false + } + + var allocCh chan uint + if len(m.freeQueues) > 0 { + var ( + found bool + bestSection uint64 + bestIndex uint + ) + for bloomIndex, _ := range m.freeQueues { + if !found || m.reqs[bloomIndex][0] < bestSection { + found = true + bestIndex = bloomIndex + bestSection = m.reqs[bloomIndex][0] + } + } + delete(m.freeQueues, bestIndex) + m.lock.Unlock() + return bestIndex, true + } else { + allocCh = make(chan uint) + m.allocQueue = append(m.allocQueue, allocCh) + } + m.lock.Unlock() + + bloomIndex, ok := <-allocCh + return bloomIndex, ok +} + +// SectionCount returns the length of the section index queue belonging to the given bloom bit index +func (m *Matcher) SectionCount(bloomIndex uint) int { + m.lock.Lock() + defer m.lock.Unlock() + + return len(m.reqs[bloomIndex]) +} + +// FetchSections fetches all or part of an already allocated queue and deallocates it +func (m *Matcher) FetchSections(bloomIndex uint, maxCount int) []uint64 { + m.lock.Lock() + defer m.lock.Unlock() + + queue := m.reqs[bloomIndex] + if maxCount < len(queue) { + // return only part of the existing queue, mark the rest as free + m.reqs[bloomIndex] = queue[maxCount:] + m.freeQueue(bloomIndex) + return queue[:maxCount] + } else { + // return the entire queue + delete(m.reqs, bloomIndex) + return queue + } +} + +// Deliver delivers a bit vector to the appropriate fetcher. +// It is possible to deliver data even after Stop has been called. Once a vector has been +// requested, the matcher will keep waiting for delivery. +func (m *Matcher) Deliver(bloomIndex uint, sectionIdxList []uint64, data [][]byte) { + m.fetchers[bloomIndex].deliver(sectionIdxList, data) +} diff --git a/core/bloombits/matcher_test.go b/core/bloombits/matcher_test.go new file mode 100644 index 000000000..bef1491b8 --- /dev/null +++ b/core/bloombits/matcher_test.go @@ -0,0 +1,196 @@ +// Copyright 2017 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 . +package bloombits + +import ( + "math/rand" + "sync/atomic" + "testing" + "time" + + "github.com/ethereum/go-ethereum/core/types" +) + +const testSectionSize = 4096 + +func matcherTestVector(b uint, s uint64) []byte { + r := make([]byte, testSectionSize/8) + for i, _ := range r { + var bb byte + for bit := 0; bit < 8; bit++ { + blockIdx := s*testSectionSize + uint64(i*8+bit) + bb += bb + if (blockIdx % uint64(b)) == 0 { + bb++ + } + } + r[i] = bb + } + return r +} + +func expMatch1(idxs types.BloomIndexList, i uint64) bool { + for _, ii := range idxs { + if (i % uint64(ii)) != 0 { + return false + } + } + return true +} + +func expMatch2(idxs []types.BloomIndexList, i uint64) bool { + for _, ii := range idxs { + if expMatch1(ii, i) { + return true + } + } + return false +} + +func expMatch3(idxs [][]types.BloomIndexList, i uint64) bool { + for _, ii := range idxs { + if !expMatch2(ii, i) { + return false + } + } + return true +} + +func testServeMatcher(m *Matcher, stop chan struct{}, cnt *uint32, maxRequestLen int) { + // serve matcher with test vectors + for i := 0; i < 10; i++ { + go func() { + for { + select { + case <-stop: + return + default: + } + b, ok := m.AllocSectionQueue() + if !ok { + return + } + if m.SectionCount(b) < maxRequestLen { + time.Sleep(time.Microsecond * 100) + } + s := m.FetchSections(b, maxRequestLen) + res := make([][]byte, len(s)) + for i, ss := range s { + res[i] = matcherTestVector(b, ss) + atomic.AddUint32(cnt, 1) + } + m.Deliver(b, s, res) + } + }() + } +} + +func testMatcher(t *testing.T, idxs [][]types.BloomIndexList, cnt uint64, stopOnMatches bool, expCount uint32) uint32 { + count1 := testMatcherWithReqCount(t, idxs, cnt, stopOnMatches, expCount, 1) + count16 := testMatcherWithReqCount(t, idxs, cnt, stopOnMatches, expCount, 16) + if count1 != count16 { + t.Errorf("Error matching idxs = %v count = %v stopOnMatches = %v: request count mismatch, %v with maxReqCount = 1 vs. %v with maxReqCount = 16", idxs, cnt, stopOnMatches, count1, count16) + } + return count1 +} + +func testMatcherWithReqCount(t *testing.T, idxs [][]types.BloomIndexList, cnt uint64, stopOnMatches bool, expCount uint32, maxReqCount int) uint32 { + m := NewMatcher(testSectionSize, nil, nil) + + for _, idxss := range idxs { + for _, idxs := range idxss { + for _, idx := range idxs { + m.newFetcher(idx) + } + } + } + + m.addresses = idxs[0] + m.topics = idxs[1:] + var reqCount uint32 + + stop := make(chan struct{}) + chn := m.Start(0, cnt-1) + testServeMatcher(m, stop, &reqCount, maxReqCount) + + for i := uint64(0); i < cnt; i++ { + if expMatch3(idxs, i) { + match, ok := <-chn + if !ok { + t.Errorf("Error matching idxs = %v count = %v stopOnMatches = %v: expected #%v, results channel closed", idxs, cnt, stopOnMatches, i) + return 0 + } + if match != i { + t.Errorf("Error matching idxs = %v count = %v stopOnMatches = %v: expected #%v, got #%v", idxs, cnt, stopOnMatches, i, match) + } + if stopOnMatches { + m.Stop() + close(stop) + stop = make(chan struct{}) + chn = m.Start(i+1, cnt-1) + testServeMatcher(m, stop, &reqCount, maxReqCount) + } + } + } + match, ok := <-chn + if ok { + t.Errorf("Error matching idxs = %v count = %v stopOnMatches = %v: expected closed channel, got #%v", idxs, cnt, stopOnMatches, match) + } + m.Stop() + close(stop) + + if expCount != 0 && expCount != reqCount { + t.Errorf("Error matching idxs = %v count = %v stopOnMatches = %v: request count mismatch, expected #%v, got #%v", idxs, cnt, stopOnMatches, expCount, reqCount) + } + + return reqCount +} + +func testRandomIdxs(l []int, max int) [][]types.BloomIndexList { + res := make([][]types.BloomIndexList, len(l)) + for i, ll := range l { + res[i] = make([]types.BloomIndexList, ll) + for j, _ := range res[i] { + for k, _ := range res[i][j] { + res[i][j][k] = uint(rand.Intn(max-1) + 2) + } + } + } + return res +} + +func TestMatcher(t *testing.T) { + testMatcher(t, [][]types.BloomIndexList{{{10, 20, 30}}}, 100000, false, 75) + testMatcher(t, [][]types.BloomIndexList{{{32, 3125, 100}}, {{40, 50, 10}}}, 100000, false, 81) + testMatcher(t, [][]types.BloomIndexList{{{4, 8, 11}, {7, 8, 17}}, {{9, 9, 12}, {15, 20, 13}}, {{18, 15, 15}, {12, 10, 4}}}, 10000, false, 36) +} + +func TestMatcherStopOnMatches(t *testing.T) { + testMatcher(t, [][]types.BloomIndexList{{{10, 20, 30}}}, 100000, true, 75) + testMatcher(t, [][]types.BloomIndexList{{{4, 8, 11}, {7, 8, 17}}, {{9, 9, 12}, {15, 20, 13}}, {{18, 15, 15}, {12, 10, 4}}}, 10000, true, 36) +} + +func TestMatcherRandom(t *testing.T) { + for i := 0; i < 20; i++ { + testMatcher(t, testRandomIdxs([]int{1}, 50), 100000, false, 0) + testMatcher(t, testRandomIdxs([]int{3}, 50), 100000, false, 0) + testMatcher(t, testRandomIdxs([]int{2, 2, 2}, 20), 100000, false, 0) + testMatcher(t, testRandomIdxs([]int{5, 5, 5}, 50), 100000, false, 0) + idxs := testRandomIdxs([]int{2, 2, 2}, 20) + reqCount := testMatcher(t, idxs, 10000, false, 0) + testMatcher(t, idxs, 10000, true, reqCount) + } +} diff --git a/core/bloombits/utils.go b/core/bloombits/utils.go new file mode 100644 index 000000000..d0755cb65 --- /dev/null +++ b/core/bloombits/utils.go @@ -0,0 +1,63 @@ +// Copyright 2017 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 . +package bloombits + +import ( + "github.com/ethereum/go-ethereum/core/types" +) + +const BloomLength = 2048 + +// BloomBitsCreator takes SectionSize number of header bloom filters and calculates the bloomBits vectors of the section +type BloomBitsCreator struct { + blooms [BloomLength][]byte + sectionSize, bitIndex uint64 +} + +func NewBloomBitsCreator(sectionSize uint64) *BloomBitsCreator { + b := &BloomBitsCreator{sectionSize: sectionSize} + for i, _ := range b.blooms { + b.blooms[i] = make([]byte, sectionSize/8) + } + return b +} + +// AddHeaderBloom takes a single bloom filter and sets the corresponding bit column in memory accordingly +func (b *BloomBitsCreator) AddHeaderBloom(bloom types.Bloom) { + if b.bitIndex >= b.sectionSize { + panic("too many header blooms added") + } + + byteIdx := b.bitIndex / 8 + bitMask := byte(1) << byte(7-b.bitIndex%8) + for bloomBitIdx, _ := range b.blooms { + bloomByteIdx := BloomLength/8 - 1 - bloomBitIdx/8 + bloomBitMask := byte(1) << byte(bloomBitIdx%8) + if (bloom[bloomByteIdx] & bloomBitMask) != 0 { + b.blooms[bloomBitIdx][byteIdx] |= bitMask + } + } + b.bitIndex++ +} + +// GetBitVector returns the bit vector belonging to the given bit index after header blooms have been added +func (b *BloomBitsCreator) GetBitVector(idx uint) []byte { + if b.bitIndex != b.sectionSize { + panic("not enough header blooms added") + } + + return b.blooms[idx][:] +} diff --git a/core/chain_indexer.go b/core/chain_indexer.go index 9a88a5b1b..56360b59a 100644 --- a/core/chain_indexer.go +++ b/core/chain_indexer.go @@ -36,7 +36,7 @@ import ( type ChainIndexerBackend interface { // Reset initiates the processing of a new chain segment, potentially terminating // any partially completed operations (in case of a reorg). - Reset(section uint64) + Reset(section uint64, lastSectionHead common.Hash) // Process crunches through the next header in the chain segment. The caller // will ensure a sequential order of headers. @@ -44,7 +44,7 @@ type ChainIndexerBackend interface { // Commit finalizes the section metadata and stores it into the database. This // interface will usually be a batch writer. - Commit(db ethdb.Database) error + Commit() error } // ChainIndexer does a post-processing job for equally sized sections of the @@ -101,10 +101,34 @@ func NewChainIndexer(chainDb, indexDb ethdb.Database, backend ChainIndexerBacken return c } +// AddKnownSectionHead marks a new section head as known/processed if it is newer +// than the already known best section head +func (c *ChainIndexer) AddKnownSectionHead(section uint64, shead common.Hash) { + c.lock.Lock() + defer c.lock.Unlock() + + if section < c.storedSections { + return + } + c.setSectionHead(section, shead) + c.setValidSections(section + 1) +} + +// IndexerChain interface is used for connecting the indexer to a blockchain +type IndexerChain interface { + CurrentHeader() *types.Header + SubscribeChainEvent(ch chan<- ChainEvent) event.Subscription +} + // Start creates a goroutine to feed chain head events into the indexer for -// cascading background processing. -func (c *ChainIndexer) Start(currentHeader *types.Header, eventMux *event.TypeMux) { - go c.eventLoop(currentHeader, eventMux) +// cascading background processing. Children do not need to be started, they +// are notified about new events by their parents. +func (c *ChainIndexer) Start(chain IndexerChain) { + ch := make(chan ChainEvent, 10) + sub := chain.SubscribeChainEvent(ch) + currentHeader := chain.CurrentHeader() + + go c.eventLoop(currentHeader, ch, sub) } // Close tears down all goroutines belonging to the indexer and returns any error @@ -125,6 +149,14 @@ func (c *ChainIndexer) Close() error { errs = append(errs, err) } } + + // Close all children + for _, child := range c.children { + if err := child.Close(); err != nil { + errs = append(errs, err) + } + } + // Return any failures switch { case len(errs) == 0: @@ -141,12 +173,10 @@ func (c *ChainIndexer) Close() error { // eventLoop is a secondary - optional - event loop of the indexer which is only // started for the outermost indexer to push chain head events into a processing // queue. -func (c *ChainIndexer) eventLoop(currentHeader *types.Header, eventMux *event.TypeMux) { +func (c *ChainIndexer) eventLoop(currentHeader *types.Header, ch chan ChainEvent, sub event.Subscription) { // Mark the chain indexer as active, requiring an additional teardown atomic.StoreUint32(&c.active, 1) - // Subscribe to chain head events - sub := eventMux.Subscribe(ChainEvent{}) defer sub.Unsubscribe() // Fire the initial new head event to start any outstanding processing @@ -163,14 +193,14 @@ func (c *ChainIndexer) eventLoop(currentHeader *types.Header, eventMux *event.Ty errc <- nil return - case ev, ok := <-sub.Chan(): + case ev, ok := <-ch: // Received a new event, ensure it's not nil (closing) and update if !ok { errc := <-c.quit errc <- nil return } - header := ev.Data.(ChainEvent).Block.Header() + header := ev.Block.Header() if header.ParentHash != prevHash { c.newHead(FindCommonAncestor(c.chainDb, prevHeader, header).Number.Uint64(), true) } @@ -226,7 +256,10 @@ func (c *ChainIndexer) newHead(head uint64, reorg bool) { // updateLoop is the main event loop of the indexer which pushes chain segments // down into the processing backend. func (c *ChainIndexer) updateLoop() { - var updated time.Time + var ( + updated time.Time + updateMsg bool + ) for { select { @@ -242,6 +275,7 @@ func (c *ChainIndexer) updateLoop() { // Periodically print an upgrade log message to the user if time.Since(updated) > 8*time.Second { if c.knownSections > c.storedSections+1 { + updateMsg = true c.log.Info("Upgrading chain index", "percentage", c.storedSections*100/c.knownSections) } updated = time.Now() @@ -250,17 +284,24 @@ func (c *ChainIndexer) updateLoop() { section := c.storedSections var oldHead common.Hash if section > 0 { - oldHead = c.sectionHead(section - 1) + oldHead = c.SectionHead(section - 1) } // Process the newly defined section in the background c.lock.Unlock() newHead, err := c.processSection(section, oldHead) + if err != nil { + c.log.Error("Section processing failed", "error", err) + } c.lock.Lock() // If processing succeeded and no reorgs occcurred, mark the section completed - if err == nil && oldHead == c.sectionHead(section-1) { + if err == nil && oldHead == c.SectionHead(section-1) { c.setSectionHead(section, newHead) c.setValidSections(section + 1) + if c.storedSections == c.knownSections && updateMsg { + updateMsg = false + c.log.Info("Finished upgrading chain index") + } c.cascadedHead = c.storedSections*c.sectionSize - 1 for _, child := range c.children { @@ -295,7 +336,7 @@ func (c *ChainIndexer) processSection(section uint64, lastHead common.Hash) (com c.log.Trace("Processing new chain section", "section", section) // Reset and partial processing - c.backend.Reset(section) + c.backend.Reset(section, lastHead) for number := section * c.sectionSize; number < (section+1)*c.sectionSize; number++ { hash := GetCanonicalHash(c.chainDb, number) @@ -311,7 +352,8 @@ func (c *ChainIndexer) processSection(section uint64, lastHead common.Hash) (com c.backend.Process(header) lastHead = header.Hash() } - if err := c.backend.Commit(c.chainDb); err != nil { + if err := c.backend.Commit(); err != nil { + c.log.Error("Section commit failed", "error", err) return common.Hash{}, err } return lastHead, nil @@ -324,7 +366,7 @@ func (c *ChainIndexer) Sections() (uint64, uint64, common.Hash) { c.lock.Lock() defer c.lock.Unlock() - return c.storedSections, c.storedSections*c.sectionSize - 1, c.sectionHead(c.storedSections - 1) + return c.storedSections, c.storedSections*c.sectionSize - 1, c.SectionHead(c.storedSections - 1) } // AddChildIndexer adds a child ChainIndexer that can use the output of this one @@ -366,7 +408,7 @@ func (c *ChainIndexer) setValidSections(sections uint64) { // sectionHead retrieves the last block hash of a processed section from the // index database. -func (c *ChainIndexer) sectionHead(section uint64) common.Hash { +func (c *ChainIndexer) SectionHead(section uint64) common.Hash { var data [8]byte binary.BigEndian.PutUint64(data[:], section) diff --git a/core/chain_indexer_test.go b/core/chain_indexer_test.go index 780e46e43..247f52cf9 100644 --- a/core/chain_indexer_test.go +++ b/core/chain_indexer_test.go @@ -23,6 +23,7 @@ import ( "testing" "time" + "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/ethdb" ) @@ -58,7 +59,6 @@ func testChainIndexer(t *testing.T, count int) { ) backends[i] = &testChainIndexBackend{t: t, processCh: make(chan uint64)} backends[i].indexer = NewChainIndexer(db, ethdb.NewTable(db, string([]byte{byte(i)})), backends[i], sectionSize, confirmsReq, 0, fmt.Sprintf("indexer-%d", i)) - defer backends[i].indexer.Close() if sections, _, _ := backends[i].indexer.Sections(); sections != 0 { t.Fatalf("Canonical section count mismatch: have %v, want %v", sections, 0) @@ -67,6 +67,7 @@ func testChainIndexer(t *testing.T, count int) { backends[i-1].indexer.AddChildIndexer(backends[i].indexer) } } + defer backends[0].indexer.Close() // parent indexer shuts down children // notify pings the root indexer about a new head or reorg, then expect // processed blocks if a section is processable notify := func(headNum, failNum uint64, reorg bool) { @@ -208,7 +209,7 @@ func (b *testChainIndexBackend) reorg(headNum uint64) uint64 { return b.stored * b.indexer.sectionSize } -func (b *testChainIndexBackend) Reset(section uint64) { +func (b *testChainIndexBackend) Reset(section uint64, lastSectionHead common.Hash) { b.section = section b.headerCnt = 0 } @@ -226,7 +227,7 @@ func (b *testChainIndexBackend) Process(header *types.Header) { } } -func (b *testChainIndexBackend) Commit(db ethdb.Database) error { +func (b *testChainIndexBackend) Commit() error { if b.headerCnt != b.indexer.sectionSize { b.t.Error("Not enough headers processed") } diff --git a/core/database_util.go b/core/database_util.go index 697111394..179d6f1b2 100644 --- a/core/database_util.go +++ b/core/database_util.go @@ -23,7 +23,6 @@ import ( "errors" "fmt" "math/big" - "sync" "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/core/types" @@ -48,9 +47,6 @@ var ( lookupPrefix = []byte("l") // lookupPrefix + hash -> transaction/receipt lookup metadata preimagePrefix = "secure-key-" // preimagePrefix + hash -> preimage - mipmapPre = []byte("mipmap-log-bloom-") - MIPMapLevels = []uint64{1000000, 500000, 100000, 50000, 1000} - configPrefix = []byte("ethereum-config-") // config prefix for the db // used by old db, now only used for conversion @@ -59,10 +55,10 @@ var ( ErrChainConfigNotFound = errors.New("ChainConfig not found") // general config not found error - mipmapBloomMu sync.Mutex // protect against race condition when updating mipmap blooms - preimageCounter = metrics.NewCounter("db/preimage/total") preimageHitCounter = metrics.NewCounter("db/preimage/hits") + + bloomBitsPrefix = []byte("bloomBits-") ) // txLookupEntry is a positional metadata to help looking up the data content of @@ -497,48 +493,6 @@ func DeleteTxLookupEntry(db ethdb.Database, hash common.Hash) { db.Delete(append(lookupPrefix, hash.Bytes()...)) } -// returns a formatted MIP mapped key by adding prefix, canonical number and level -// -// ex. fn(98, 1000) = (prefix || 1000 || 0) -func mipmapKey(num, level uint64) []byte { - lkey := make([]byte, 8) - binary.BigEndian.PutUint64(lkey, level) - key := new(big.Int).SetUint64(num / level * level) - - return append(mipmapPre, append(lkey, key.Bytes()...)...) -} - -// WriteMipmapBloom writes each address included in the receipts' logs to the -// MIP bloom bin. -func WriteMipmapBloom(db ethdb.Database, number uint64, receipts types.Receipts) error { - mipmapBloomMu.Lock() - defer mipmapBloomMu.Unlock() - - batch := db.NewBatch() - for _, level := range MIPMapLevels { - key := mipmapKey(number, level) - bloomDat, _ := db.Get(key) - bloom := types.BytesToBloom(bloomDat) - for _, receipt := range receipts { - for _, log := range receipt.Logs { - bloom.Add(log.Address.Big()) - } - } - batch.Put(key, bloom.Bytes()) - } - if err := batch.Write(); err != nil { - return fmt.Errorf("mipmap write fail for: %d: %v", number, err) - } - return nil -} - -// GetMipmapBloom returns a bloom filter using the number and level as input -// parameters. For available levels see MIPMapLevels. -func GetMipmapBloom(db ethdb.Database, number, level uint64) types.Bloom { - bloomDat, _ := db.Get(mipmapKey(number, level)) - return types.BytesToBloom(bloomDat) -} - // PreimageTable returns a Database instance with the key prefix for preimage entries. func PreimageTable(db ethdb.Database) ethdb.Database { return ethdb.NewTable(db, preimagePrefix) @@ -637,3 +591,22 @@ func FindCommonAncestor(db ethdb.Database, a, b *types.Header) *types.Header { } return a } + +// GetBloomBits reads the compressed bloomBits vector belonging to the given section and bit index from the db +func GetBloomBits(db ethdb.Database, bitIdx, sectionIdx uint64, sectionHead common.Hash) ([]byte, error) { + var encKey [10]byte + binary.BigEndian.PutUint16(encKey[0:2], uint16(bitIdx)) + binary.BigEndian.PutUint64(encKey[2:10], sectionIdx) + key := append(append(bloomBitsPrefix, encKey[:]...), sectionHead.Bytes()...) + bloomBits, err := db.Get(key) + return bloomBits, err +} + +// StoreBloomBits writes the compressed bloomBits vector belonging to the given section and bit index to the db +func StoreBloomBits(db ethdb.Database, bitIdx, sectionIdx uint64, sectionHead common.Hash, bloomBits []byte) { + var encKey [10]byte + binary.BigEndian.PutUint16(encKey[0:2], uint16(bitIdx)) + binary.BigEndian.PutUint64(encKey[2:10], sectionIdx) + key := append(append(bloomBitsPrefix, encKey[:]...), sectionHead.Bytes()...) + db.Put(key, bloomBits) +} diff --git a/core/database_util_test.go b/core/database_util_test.go index e91f1b593..940221a29 100644 --- a/core/database_util_test.go +++ b/core/database_util_test.go @@ -18,17 +18,13 @@ package core import ( "bytes" - "io/ioutil" "math/big" - "os" "testing" "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/core/types" - "github.com/ethereum/go-ethereum/crypto" "github.com/ethereum/go-ethereum/crypto/sha3" "github.com/ethereum/go-ethereum/ethdb" - "github.com/ethereum/go-ethereum/params" "github.com/ethereum/go-ethereum/rlp" ) @@ -390,107 +386,3 @@ func TestBlockReceiptStorage(t *testing.T) { t.Fatalf("deleted receipts returned: %v", rs) } } - -func TestMipmapBloom(t *testing.T) { - db, _ := ethdb.NewMemDatabase() - - receipt1 := new(types.Receipt) - receipt1.Logs = []*types.Log{ - {Address: common.BytesToAddress([]byte("test"))}, - {Address: common.BytesToAddress([]byte("address"))}, - } - receipt2 := new(types.Receipt) - receipt2.Logs = []*types.Log{ - {Address: common.BytesToAddress([]byte("test"))}, - {Address: common.BytesToAddress([]byte("address1"))}, - } - - WriteMipmapBloom(db, 1, types.Receipts{receipt1}) - WriteMipmapBloom(db, 2, types.Receipts{receipt2}) - - for _, level := range MIPMapLevels { - bloom := GetMipmapBloom(db, 2, level) - if !bloom.Test(new(big.Int).SetBytes([]byte("address1"))) { - t.Error("expected test to be included on level:", level) - } - } - - // reset - db, _ = ethdb.NewMemDatabase() - receipt := new(types.Receipt) - receipt.Logs = []*types.Log{ - {Address: common.BytesToAddress([]byte("test"))}, - } - WriteMipmapBloom(db, 999, types.Receipts{receipt1}) - - receipt = new(types.Receipt) - receipt.Logs = []*types.Log{ - {Address: common.BytesToAddress([]byte("test 1"))}, - } - WriteMipmapBloom(db, 1000, types.Receipts{receipt}) - - bloom := GetMipmapBloom(db, 1000, 1000) - if bloom.TestBytes([]byte("test")) { - t.Error("test should not have been included") - } -} - -func TestMipmapChain(t *testing.T) { - dir, err := ioutil.TempDir("", "mipmap") - if err != nil { - t.Fatal(err) - } - defer os.RemoveAll(dir) - - var ( - db, _ = ethdb.NewLDBDatabase(dir, 0, 0) - key1, _ = crypto.HexToECDSA("b71c71a67e1177ad4e901695e1b4b9ee17ae16c6668d313eac2f96dbcda3f291") - addr = crypto.PubkeyToAddress(key1.PublicKey) - addr2 = common.BytesToAddress([]byte("jeff")) - - hash1 = common.BytesToHash([]byte("topic1")) - ) - defer db.Close() - - gspec := &Genesis{ - Config: params.TestChainConfig, - Alloc: GenesisAlloc{addr: {Balance: big.NewInt(1000000)}}, - } - genesis := gspec.MustCommit(db) - chain, receipts := GenerateChain(params.TestChainConfig, genesis, db, 1010, func(i int, gen *BlockGen) { - var receipts types.Receipts - switch i { - case 1: - receipt := types.NewReceipt(nil, false, new(big.Int)) - receipt.Logs = []*types.Log{{Address: addr, Topics: []common.Hash{hash1}}} - gen.AddUncheckedReceipt(receipt) - receipts = types.Receipts{receipt} - case 1000: - receipt := types.NewReceipt(nil, false, new(big.Int)) - receipt.Logs = []*types.Log{{Address: addr2}} - gen.AddUncheckedReceipt(receipt) - receipts = types.Receipts{receipt} - - } - - // store the receipts - WriteMipmapBloom(db, uint64(i+1), receipts) - }) - for i, block := range chain { - WriteBlock(db, block) - if err := WriteCanonicalHash(db, block.Hash(), block.NumberU64()); err != nil { - t.Fatalf("failed to insert block number: %v", err) - } - if err := WriteHeadBlockHash(db, block.Hash()); err != nil { - t.Fatalf("failed to insert block number: %v", err) - } - if err := WriteBlockReceipts(db, block.Hash(), block.NumberU64(), receipts[i]); err != nil { - t.Fatal("error writing block receipts:", err) - } - } - - bloom := GetMipmapBloom(db, 0, 1000) - if bloom.TestBytes(addr2[:]) { - t.Error("address was included in bloom and should not have") - } -} diff --git a/core/types/bloom9.go b/core/types/bloom9.go index 60aacc301..bdc6e60e7 100644 --- a/core/types/bloom9.go +++ b/core/types/bloom9.go @@ -106,6 +106,20 @@ func LogsBloom(logs []*Log) *big.Int { return bin } +type BloomIndexList [3]uint + +// BloomIndexes returns the bloom filter bit indexes belonging to the given key +func BloomIndexes(b []byte) BloomIndexList { + b = crypto.Keccak256(b[:]) + + var r [3]uint + for i, _ := range r { + r[i] = (uint(b[i+i+1]) + (uint(b[i+i]) << 8)) & 2047 + } + + return r +} + func bloom9(b []byte) *big.Int { b = crypto.Keccak256(b[:]) diff --git a/eth/api_backend.go b/eth/api_backend.go index 19ef79f23..fa3cf3f80 100644 --- a/eth/api_backend.go +++ b/eth/api_backend.go @@ -28,6 +28,7 @@ import ( "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/core/vm" "github.com/ethereum/go-ethereum/eth/downloader" + "github.com/ethereum/go-ethereum/eth/filters" "github.com/ethereum/go-ethereum/eth/gasprice" "github.com/ethereum/go-ethereum/ethdb" "github.com/ethereum/go-ethereum/event" @@ -194,3 +195,29 @@ func (b *EthApiBackend) EventMux() *event.TypeMux { func (b *EthApiBackend) AccountManager() *accounts.Manager { return b.eth.AccountManager() } + +func (b *EthApiBackend) GetBloomBits(ctx context.Context, bitIdx uint64, sectionIdxList []uint64) ([][]byte, error) { + results := make([][]byte, len(sectionIdxList)) + var err error + for i, sectionIdx := range sectionIdxList { + sectionHead := core.GetCanonicalHash(b.eth.chainDb, (sectionIdx+1)*bloomBitsSection-1) + results[i], err = core.GetBloomBits(b.eth.chainDb, bitIdx, sectionIdx, sectionHead) + if err != nil { + return nil, err + } + } + return results, nil +} + +func (b *EthApiBackend) BloomBitsSections() uint64 { + sections, _, _ := b.eth.bbIndexer.Sections() + return sections +} + +func (b *EthApiBackend) BloomBitsConfig() filters.BloomConfig { + return filters.BloomConfig{ + SectionSize: bloomBitsSection, + MaxRequestLen: 16, + MaxRequestWait: 0, + } +} diff --git a/eth/backend.go b/eth/backend.go index 5837c8564..efc0a2317 100644 --- a/eth/backend.go +++ b/eth/backend.go @@ -77,6 +77,8 @@ type Ethereum struct { engine consensus.Engine accountManager *accounts.Manager + bbIndexer *core.ChainIndexer + ApiBackend *EthApiBackend miner *miner.Miner @@ -125,11 +127,9 @@ func New(ctx *node.ServiceContext, config *Config) (*Ethereum, error) { networkId: config.NetworkId, gasPrice: config.GasPrice, etherbase: config.Etherbase, + bbIndexer: NewBloomBitsProcessor(chainDb, bloomBitsSection), } - if err := addMipmapBloomBins(chainDb); err != nil { - return nil, err - } log.Info("Initialising Ethereum protocol", "versions", ProtocolVersions, "network", config.NetworkId) if !config.SkipBcVersionCheck { @@ -151,6 +151,7 @@ func New(ctx *node.ServiceContext, config *Config) (*Ethereum, error) { eth.blockchain.SetHead(compat.RewindTo) core.WriteChainConfig(chainDb, genesisHash, chainConfig) } + eth.bbIndexer.Start(eth.blockchain) if config.TxPool.Journal != "" { config.TxPool.Journal = ctx.ResolvePath(config.TxPool.Journal) @@ -260,7 +261,7 @@ func (s *Ethereum) APIs() []rpc.API { }, { Namespace: "eth", Version: "1.0", - Service: filters.NewPublicFilterAPI(s.ApiBackend, false), + Service: filters.NewPublicFilterAPI(s.ApiBackend, false, bloomBitsSection), Public: true, }, { Namespace: "admin", @@ -389,6 +390,7 @@ func (s *Ethereum) Stop() error { if s.stopDbUpgrade != nil { s.stopDbUpgrade() } + s.bbIndexer.Close() s.blockchain.Stop() s.protocolManager.Stop() if s.lesServer != nil { diff --git a/eth/backend_test.go b/eth/backend_test.go deleted file mode 100644 index 1fd25e95a..000000000 --- a/eth/backend_test.go +++ /dev/null @@ -1,74 +0,0 @@ -// Copyright 2015 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 . - -package eth - -import ( - "math/big" - "testing" - - "github.com/ethereum/go-ethereum/common" - "github.com/ethereum/go-ethereum/core" - "github.com/ethereum/go-ethereum/core/types" - "github.com/ethereum/go-ethereum/ethdb" - "github.com/ethereum/go-ethereum/params" -) - -func TestMipmapUpgrade(t *testing.T) { - db, _ := ethdb.NewMemDatabase() - addr := common.BytesToAddress([]byte("jeff")) - genesis := new(core.Genesis).MustCommit(db) - - chain, receipts := core.GenerateChain(params.TestChainConfig, genesis, db, 10, func(i int, gen *core.BlockGen) { - switch i { - case 1: - receipt := types.NewReceipt(nil, false, new(big.Int)) - receipt.Logs = []*types.Log{{Address: addr}} - gen.AddUncheckedReceipt(receipt) - case 2: - receipt := types.NewReceipt(nil, false, new(big.Int)) - receipt.Logs = []*types.Log{{Address: addr}} - gen.AddUncheckedReceipt(receipt) - } - }) - for i, block := range chain { - core.WriteBlock(db, block) - if err := core.WriteCanonicalHash(db, block.Hash(), block.NumberU64()); err != nil { - t.Fatalf("failed to insert block number: %v", err) - } - if err := core.WriteHeadBlockHash(db, block.Hash()); err != nil { - t.Fatalf("failed to insert block number: %v", err) - } - if err := core.WriteBlockReceipts(db, block.Hash(), block.NumberU64(), receipts[i]); err != nil { - t.Fatal("error writing block receipts:", err) - } - } - - err := addMipmapBloomBins(db) - if err != nil { - t.Fatal(err) - } - - bloom := core.GetMipmapBloom(db, 1, core.MIPMapLevels[0]) - if (bloom == types.Bloom{}) { - t.Error("got empty bloom filter") - } - - data, _ := db.Get([]byte("setting-mipmap-version")) - if len(data) == 0 { - t.Error("setting-mipmap-version not written to database") - } -} diff --git a/eth/db_upgrade.go b/eth/db_upgrade.go index 90111b2b3..ce8ce699a 100644 --- a/eth/db_upgrade.go +++ b/eth/db_upgrade.go @@ -19,11 +19,13 @@ package eth import ( "bytes" - "fmt" "time" "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/common/bitutil" "github.com/ethereum/go-ethereum/core" + "github.com/ethereum/go-ethereum/core/bloombits" + "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/ethdb" "github.com/ethereum/go-ethereum/log" "github.com/ethereum/go-ethereum/rlp" @@ -135,45 +137,37 @@ func upgradeDeduplicateData(db ethdb.Database) func() error { } } -func addMipmapBloomBins(db ethdb.Database) (err error) { - const mipmapVersion uint = 2 +// BloomBitsIndex implements ChainIndex +type BloomBitsIndex struct { + db ethdb.Database + bc *bloombits.BloomBitsCreator + section, sectionSize uint64 + sectionHead common.Hash +} - // check if the version is set. We ignore data for now since there's - // only one version so we can easily ignore it for now - var data []byte - data, _ = db.Get([]byte("setting-mipmap-version")) - if len(data) > 0 { - var version uint - if err := rlp.DecodeBytes(data, &version); err == nil && version == mipmapVersion { - return nil - } - } +// number of confirmation blocks before a section is considered probably final and its bloom bits are calculated +const bloomBitsConfirmations = 256 - defer func() { - if err == nil { - var val []byte - val, err = rlp.EncodeToBytes(mipmapVersion) - if err == nil { - err = db.Put([]byte("setting-mipmap-version"), val) - } - return - } - }() - latestHash := core.GetHeadBlockHash(db) - latestBlock := core.GetBlock(db, latestHash, core.GetBlockNumber(db, latestHash)) - if latestBlock == nil { // clean database - return - } +// NewBloomBitsProcessor returns a chain processor that generates bloom bits data for the canonical chain +func NewBloomBitsProcessor(db ethdb.Database, sectionSize uint64) *core.ChainIndexer { + backend := &BloomBitsIndex{db: db, sectionSize: sectionSize} + return core.NewChainIndexer(db, ethdb.NewTable(db, "bbIndex-"), backend, sectionSize, bloomBitsConfirmations, time.Millisecond*100, "bloombits") +} - tstart := time.Now() - log.Warn("Upgrading db log bloom bins") - for i := uint64(0); i <= latestBlock.NumberU64(); i++ { - hash := core.GetCanonicalHash(db, i) - if (hash == common.Hash{}) { - return fmt.Errorf("chain db corrupted. Could not find block %d.", i) - } - core.WriteMipmapBloom(db, i, core.GetBlockReceipts(db, hash, i)) +func (b *BloomBitsIndex) Reset(section uint64, lastSectionHead common.Hash) { + b.bc = bloombits.NewBloomBitsCreator(b.sectionSize) + b.section = section +} + +func (b *BloomBitsIndex) Process(header *types.Header) { + b.bc.AddHeaderBloom(header.Bloom) + b.sectionHead = header.Hash() +} + +func (b *BloomBitsIndex) Commit() error { + for i := 0; i < bloombits.BloomLength; i++ { + compVector := bitutil.CompressBytes(b.bc.GetBitVector(uint(i))) + core.StoreBloomBits(b.db, uint64(i), b.section, b.sectionHead, compVector) } - log.Info("Bloom-bin upgrade completed", "elapsed", common.PrettyDuration(time.Since(tstart))) return nil } diff --git a/eth/filters/api.go b/eth/filters/api.go index fff58a268..11767753e 100644 --- a/eth/filters/api.go +++ b/eth/filters/api.go @@ -51,24 +51,25 @@ type filter struct { // PublicFilterAPI offers support to create and manage filters. This will allow external clients to retrieve various // information related to the Ethereum protocol such als blocks, transactions and logs. type PublicFilterAPI struct { - backend Backend - useMipMap bool - mux *event.TypeMux - chainDb ethdb.Database - events *EventSystem - filtersMu sync.Mutex - filters map[rpc.ID]*filter + backend Backend + bloomBitsSection uint64 + mux *event.TypeMux + quit chan struct{} + chainDb ethdb.Database + events *EventSystem + filtersMu sync.Mutex + filters map[rpc.ID]*filter } // NewPublicFilterAPI returns a new PublicFilterAPI instance. -func NewPublicFilterAPI(backend Backend, lightMode bool) *PublicFilterAPI { +func NewPublicFilterAPI(backend Backend, lightMode bool, bloomBitsSection uint64) *PublicFilterAPI { api := &PublicFilterAPI{ - backend: backend, - useMipMap: !lightMode, - mux: backend.EventMux(), - chainDb: backend.ChainDb(), - events: NewEventSystem(backend.EventMux(), backend, lightMode), - filters: make(map[rpc.ID]*filter), + backend: backend, + bloomBitsSection: bloomBitsSection, + mux: backend.EventMux(), + chainDb: backend.ChainDb(), + events: NewEventSystem(backend.EventMux(), backend, lightMode), + filters: make(map[rpc.ID]*filter), } go api.timeoutLoop() @@ -332,11 +333,7 @@ func (api *PublicFilterAPI) GetLogs(ctx context.Context, crit FilterCriteria) ([ crit.ToBlock = big.NewInt(rpc.LatestBlockNumber.Int64()) } - filter := New(api.backend, api.useMipMap) - filter.SetBeginBlock(crit.FromBlock.Int64()) - filter.SetEndBlock(crit.ToBlock.Int64()) - filter.SetAddresses(crit.Addresses) - filter.SetTopics(crit.Topics) + filter := New(api.backend, crit.FromBlock.Int64(), crit.ToBlock.Int64(), crit.Addresses, crit.Topics) logs, err := filter.Find(ctx) return returnLogs(logs), err @@ -372,19 +369,18 @@ func (api *PublicFilterAPI) GetFilterLogs(ctx context.Context, id rpc.ID) ([]*ty return nil, fmt.Errorf("filter not found") } - filter := New(api.backend, api.useMipMap) + var begin, end int64 if f.crit.FromBlock != nil { - filter.SetBeginBlock(f.crit.FromBlock.Int64()) + begin = f.crit.FromBlock.Int64() } else { - filter.SetBeginBlock(rpc.LatestBlockNumber.Int64()) + begin = rpc.LatestBlockNumber.Int64() } if f.crit.ToBlock != nil { - filter.SetEndBlock(f.crit.ToBlock.Int64()) + end = f.crit.ToBlock.Int64() } else { - filter.SetEndBlock(rpc.LatestBlockNumber.Int64()) + end = rpc.LatestBlockNumber.Int64() } - filter.SetAddresses(f.crit.Addresses) - filter.SetTopics(f.crit.Topics) + filter := New(api.backend, begin, end, f.crit.Addresses, f.crit.Topics) logs, err := filter.Find(ctx) if err != nil { diff --git a/eth/filters/bench_test.go b/eth/filters/bench_test.go new file mode 100644 index 000000000..2487bc0eb --- /dev/null +++ b/eth/filters/bench_test.go @@ -0,0 +1,237 @@ +// Copyright 2017 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 . + +package filters + +import ( + "bytes" + "context" + "fmt" + "testing" + "time" + + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/common/bitutil" + "github.com/ethereum/go-ethereum/core" + "github.com/ethereum/go-ethereum/core/bloombits" + "github.com/ethereum/go-ethereum/core/types" + "github.com/ethereum/go-ethereum/ethdb" + "github.com/ethereum/go-ethereum/event" + "github.com/ethereum/go-ethereum/node" + "github.com/golang/snappy" +) + +func BenchmarkBloomBits512(b *testing.B) { + benchmarkBloomBitsForSize(b, 512) +} + +func BenchmarkBloomBits1k(b *testing.B) { + benchmarkBloomBitsForSize(b, 1024) +} + +func BenchmarkBloomBits2k(b *testing.B) { + benchmarkBloomBitsForSize(b, 2048) +} + +func BenchmarkBloomBits4k(b *testing.B) { + benchmarkBloomBitsForSize(b, 4096) +} + +func BenchmarkBloomBits8k(b *testing.B) { + benchmarkBloomBitsForSize(b, 8192) +} + +func BenchmarkBloomBits16k(b *testing.B) { + benchmarkBloomBitsForSize(b, 16384) +} + +func BenchmarkBloomBits32k(b *testing.B) { + benchmarkBloomBitsForSize(b, 32768) +} + +func benchmarkBloomBitsForSize(b *testing.B, sectionSize uint64) { + benchmarkBloomBits(b, sectionSize, 0) + benchmarkBloomBits(b, sectionSize, 1) + benchmarkBloomBits(b, sectionSize, 2) +} + +const benchFilterCnt = 2000 + +func benchmarkBloomBits(b *testing.B, sectionSize uint64, comp int) { + benchDataDir := node.DefaultDataDir() + "/geth/chaindata" + fmt.Println("Running bloombits benchmark section size:", sectionSize, " compression method:", comp) + + var ( + compressFn func([]byte) []byte + decompressFn func([]byte, int) ([]byte, error) + ) + switch comp { + case 0: + // no compression + compressFn = func(data []byte) []byte { + return data + } + decompressFn = func(data []byte, target int) ([]byte, error) { + if len(data) != target { + panic(nil) + } + return data, nil + } + case 1: + // bitutil/compress.go + compressFn = bitutil.CompressBytes + decompressFn = bitutil.DecompressBytes + case 2: + // go snappy + compressFn = func(data []byte) []byte { + return snappy.Encode(nil, data) + } + decompressFn = func(data []byte, target int) ([]byte, error) { + decomp, err := snappy.Decode(nil, data) + if err != nil || len(decomp) != target { + panic(err) + } + return decomp, nil + } + } + + db, err := ethdb.NewLDBDatabase(benchDataDir, 128, 1024) + if err != nil { + b.Fatalf("error opening database at %v: %v", benchDataDir, err) + } + head := core.GetHeadBlockHash(db) + if head == (common.Hash{}) { + b.Fatalf("chain data not found at %v", benchDataDir) + } + + clearBloomBits(db) + fmt.Println("Generating bloombits data...") + headNum := core.GetBlockNumber(db, head) + if headNum < sectionSize+512 { + b.Fatalf("not enough blocks for running a benchmark") + } + + start := time.Now() + cnt := (headNum - 512) / sectionSize + var dataSize, compSize uint64 + for sectionIdx := uint64(0); sectionIdx < cnt; sectionIdx++ { + bc := bloombits.NewBloomBitsCreator(sectionSize) + var header *types.Header + for i := sectionIdx * sectionSize; i < (sectionIdx+1)*sectionSize; i++ { + hash := core.GetCanonicalHash(db, i) + header = core.GetHeader(db, hash, i) + if header == nil { + b.Fatalf("Error creating bloomBits data") + } + bc.AddHeaderBloom(header.Bloom) + } + sectionHead := core.GetCanonicalHash(db, (sectionIdx+1)*sectionSize-1) + for i := 0; i < bloombits.BloomLength; i++ { + data := bc.GetBitVector(uint(i)) + comp := compressFn(data) + dataSize += uint64(len(data)) + compSize += uint64(len(comp)) + core.StoreBloomBits(db, uint64(i), sectionIdx, sectionHead, comp) + } + //if sectionIdx%50 == 0 { + // fmt.Println(" section", sectionIdx, "/", cnt) + //} + } + + d := time.Since(start) + fmt.Println("Finished generating bloombits data") + fmt.Println(" ", d, "total ", d/time.Duration(cnt*sectionSize), "per block") + fmt.Println(" data size:", dataSize, " compressed size:", compSize, " compression ratio:", float64(compSize)/float64(dataSize)) + + fmt.Println("Running filter benchmarks...") + start = time.Now() + mux := new(event.TypeMux) + var backend *testBackend + + for i := 0; i < benchFilterCnt; i++ { + if i%20 == 0 { + db.Close() + db, _ = ethdb.NewLDBDatabase(benchDataDir, 128, 1024) + backend = &testBackend{mux, db, cnt, new(event.Feed), new(event.Feed), new(event.Feed), new(event.Feed)} + } + var addr common.Address + addr[0] = byte(i) + addr[1] = byte(i / 256) + filter := New(backend, 0, int64(cnt*sectionSize-1), []common.Address{addr}, nil) + filter.decompress = decompressFn + if _, err := filter.Find(context.Background()); err != nil { + b.Error("filter.Find error:", err) + } + } + d = time.Since(start) + fmt.Println("Finished running filter benchmarks") + fmt.Println(" ", d, "total ", d/time.Duration(benchFilterCnt), "per address", d*time.Duration(1000000)/time.Duration(benchFilterCnt*cnt*sectionSize), "per million blocks") + db.Close() +} + +func forEachKey(db ethdb.Database, startPrefix, endPrefix []byte, fn func(key []byte)) { + it := db.(*ethdb.LDBDatabase).NewIterator() + it.Seek(startPrefix) + for it.Valid() { + key := it.Key() + cmpLen := len(key) + if len(endPrefix) < cmpLen { + cmpLen = len(endPrefix) + } + if bytes.Compare(key[:cmpLen], endPrefix) == 1 { + break + } + fn(common.CopyBytes(key)) + it.Next() + } + it.Release() +} + +var bloomBitsPrefix = []byte("bloomBits-") + +func clearBloomBits(db ethdb.Database) { + fmt.Println("Clearing bloombits data...") + forEachKey(db, bloomBitsPrefix, bloomBitsPrefix, func(key []byte) { + db.Delete(key) + }) +} + +func BenchmarkNoBloomBits(b *testing.B) { + benchDataDir := node.DefaultDataDir() + "/geth/chaindata" + fmt.Println("Running benchmark without bloombits") + db, err := ethdb.NewLDBDatabase(benchDataDir, 128, 1024) + if err != nil { + b.Fatalf("error opening database at %v: %v", benchDataDir, err) + } + head := core.GetHeadBlockHash(db) + if head == (common.Hash{}) { + b.Fatalf("chain data not found at %v", benchDataDir) + } + headNum := core.GetBlockNumber(db, head) + + clearBloomBits(db) + + fmt.Println("Running filter benchmarks...") + start := time.Now() + mux := new(event.TypeMux) + backend := &testBackend{mux, db, 0, new(event.Feed), new(event.Feed), new(event.Feed), new(event.Feed)} + filter := New(backend, 0, int64(headNum), []common.Address{common.Address{}}, nil) + filter.Find(context.Background()) + d := time.Since(start) + fmt.Println("Finished running filter benchmarks") + fmt.Println(" ", d, "total ", d*time.Duration(1000000)/time.Duration(headNum+1), "per million blocks") + db.Close() +} diff --git a/eth/filters/filter.go b/eth/filters/filter.go index f848bc6af..ea9ccf2f9 100644 --- a/eth/filters/filter.go +++ b/eth/filters/filter.go @@ -18,11 +18,14 @@ package filters import ( "context" - "math" "math/big" + "sync" + "time" "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/common/bitutil" "github.com/ethereum/go-ethereum/core" + "github.com/ethereum/go-ethereum/core/bloombits" "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/ethdb" "github.com/ethereum/go-ethereum/event" @@ -34,58 +37,51 @@ type Backend interface { EventMux() *event.TypeMux HeaderByNumber(ctx context.Context, blockNr rpc.BlockNumber) (*types.Header, error) GetReceipts(ctx context.Context, blockHash common.Hash) (types.Receipts, error) + BloomBitsSections() uint64 + BloomBitsConfig() BloomConfig SubscribeTxPreEvent(chan<- core.TxPreEvent) event.Subscription SubscribeChainEvent(ch chan<- core.ChainEvent) event.Subscription SubscribeRemovedLogsEvent(ch chan<- core.RemovedLogsEvent) event.Subscription SubscribeLogsEvent(ch chan<- []*types.Log) event.Subscription + GetBloomBits(ctx context.Context, bitIdx uint64, sectionIdxList []uint64) ([][]byte, error) +} + +type BloomConfig struct { + SectionSize uint64 + MaxRequestLen int + MaxRequestWait time.Duration } // Filter can be used to retrieve and filter logs. type Filter struct { - backend Backend - useMipMap bool + backend Backend + bloomBitsConfig BloomConfig db ethdb.Database begin, end int64 addresses []common.Address topics [][]common.Hash + + decompress func([]byte, int) ([]byte, error) + matcher *bloombits.Matcher } // New creates a new filter which uses a bloom filter on blocks to figure out whether // a particular block is interesting or not. -// MipMaps allow past blocks to be searched much more efficiently, but are not available -// to light clients. -func New(backend Backend, useMipMap bool) *Filter { +func New(backend Backend, begin, end int64, addresses []common.Address, topics [][]common.Hash) *Filter { return &Filter{ - backend: backend, - useMipMap: useMipMap, - db: backend.ChainDb(), + backend: backend, + begin: begin, + end: end, + addresses: addresses, + topics: topics, + bloomBitsConfig: backend.BloomBitsConfig(), + db: backend.ChainDb(), + matcher: bloombits.NewMatcher(backend.BloomBitsConfig().SectionSize, addresses, topics), + decompress: bitutil.DecompressBytes, } } -// SetBeginBlock sets the earliest block for filtering. -// -1 = latest block (i.e., the current block) -// hash = particular hash from-to -func (f *Filter) SetBeginBlock(begin int64) { - f.begin = begin -} - -// SetEndBlock sets the latest block for filtering. -func (f *Filter) SetEndBlock(end int64) { - f.end = end -} - -// SetAddresses matches only logs that are generated from addresses that are included -// in the given addresses. -func (f *Filter) SetAddresses(addr []common.Address) { - f.addresses = addr -} - -// SetTopics matches only logs that have topics matching the given topics. -func (f *Filter) SetTopics(topics [][]common.Hash) { - f.topics = topics -} - // FindOnce searches the blockchain for matching log entries, returning // all matching entries from the first block that contains matches, // updating the start point of the filter accordingly. If no results are @@ -106,18 +102,9 @@ func (f *Filter) FindOnce(ctx context.Context) ([]*types.Log, error) { endBlockNo = headBlockNumber } - // if no addresses are present we can't make use of fast search which - // uses the mipmap bloom filters to check for fast inclusion and uses - // higher range probability in order to ensure at least a false positive - if !f.useMipMap || len(f.addresses) == 0 { - logs, blockNumber, err := f.getLogs(ctx, beginBlockNo, endBlockNo) - f.begin = int64(blockNumber + 1) - return logs, err - } - - logs, blockNumber := f.mipFind(beginBlockNo, endBlockNo, 0) + logs, blockNumber, err := f.getLogs(ctx, beginBlockNo, endBlockNo) f.begin = int64(blockNumber + 1) - return logs, nil + return logs, err } // Run filters logs with the current parameters set @@ -131,43 +118,134 @@ func (f *Filter) Find(ctx context.Context) (logs []*types.Log, err error) { } } -func (f *Filter) mipFind(start, end uint64, depth int) (logs []*types.Log, blockNumber uint64) { - level := core.MIPMapLevels[depth] - // normalise numerator so we can work in level specific batches and - // work with the proper range checks - for num := start / level * level; num <= end; num += level { - // find addresses in bloom filters - bloom := core.GetMipmapBloom(f.db, num, level) - // Don't bother checking the first time through the loop - we're probably picking - // up where a previous run left off. - first := true - for _, addr := range f.addresses { - if first || bloom.TestBytes(addr[:]) { - first = false - // range check normalised values and make sure that - // we're resolving the correct range instead of the - // normalised values. - start := uint64(math.Max(float64(num), float64(start))) - end := uint64(math.Min(float64(num+level-1), float64(end))) - if depth+1 == len(core.MIPMapLevels) { - l, blockNumber, _ := f.getLogs(context.Background(), start, end) - if len(l) > 0 { - return l, blockNumber +// nextRequest returns the next request to retrieve for the bloombits matcher +func (f *Filter) nextRequest() (bloombits uint, sections []uint64) { + bloomIndex, ok := f.matcher.AllocSectionQueue() + if !ok { + return 0, nil + } + if f.bloomBitsConfig.MaxRequestWait > 0 && + (f.bloomBitsConfig.MaxRequestLen <= 1 || // SectionCount is always greater than zero after a successful alloc + f.matcher.SectionCount(bloomIndex) < f.bloomBitsConfig.MaxRequestLen) { + time.Sleep(f.bloomBitsConfig.MaxRequestWait) + } + return bloomIndex, f.matcher.FetchSections(bloomIndex, f.bloomBitsConfig.MaxRequestLen) +} + +// serveMatcher serves the bloombits matcher by fetching the requested vectors +// through the filter backend +func (f *Filter) serveMatcher(ctx context.Context, stop chan struct{}, wg *sync.WaitGroup) chan error { + errChn := make(chan error, 1) + wg.Add(10) + for i := 0; i < 10; i++ { + go func(i int) { + defer wg.Done() + + for { + b, s := f.nextRequest() + if s == nil { + return + } + data, err := f.backend.GetBloomBits(ctx, uint64(b), s) + if err != nil { + select { + case errChn <- err: + case <-stop: } - } else { - l, blockNumber := f.mipFind(start, end, depth+1) - if len(l) > 0 { - return l, blockNumber + return + } + decomp := make([][]byte, len(data)) + for i, d := range data { + var err error + if decomp[i], err = f.decompress(d, int(f.bloomBitsConfig.SectionSize/8)); err != nil { + select { + case errChn <- err: + case <-stop: + } + return } } + f.matcher.Deliver(b, s, decomp) } - } + }(i) } - return nil, end + return errChn +} + +// checkMatches checks if the receipts belonging to the given header contain any log events that +// match the filter criteria. This function is called when the bloom filter signals a potential match. +func (f *Filter) checkMatches(ctx context.Context, header *types.Header) (logs []*types.Log, err error) { + // Get the logs of the block + receipts, err := f.backend.GetReceipts(ctx, header.Hash()) + if err != nil { + return nil, err + } + var unfiltered []*types.Log + for _, receipt := range receipts { + unfiltered = append(unfiltered, ([]*types.Log)(receipt.Logs)...) + } + logs = filterLogs(unfiltered, nil, nil, f.addresses, f.topics) + if len(logs) > 0 { + return logs, nil + } + return nil, nil } func (f *Filter) getLogs(ctx context.Context, start, end uint64) (logs []*types.Log, blockNumber uint64, err error) { + haveBloomBitsBefore := f.backend.BloomBitsSections() * f.bloomBitsConfig.SectionSize + if haveBloomBitsBefore > start { + e := end + if haveBloomBitsBefore <= e { + e = haveBloomBitsBefore - 1 + } + + stop := make(chan struct{}) + var wg sync.WaitGroup + matches := f.matcher.Start(start, e) + errChn := f.serveMatcher(ctx, stop, &wg) + + defer func() { + f.matcher.Stop() + close(stop) + wg.Wait() + }() + + loop: + for { + select { + case i, ok := <-matches: + if !ok { + break loop + } + + blockNumber := rpc.BlockNumber(i) + header, err := f.backend.HeaderByNumber(ctx, blockNumber) + if header == nil || err != nil { + return logs, end, err + } + + logs, err := f.checkMatches(ctx, header) + if err != nil { + return nil, end, err + } + if logs != nil { + return logs, i, nil + } + case err := <-errChn: + return logs, end, err + case <-ctx.Done(): + return nil, end, ctx.Err() + } + } + + if end < haveBloomBitsBefore { + return logs, end, nil + } + start = haveBloomBitsBefore + } + + // search the rest with regular block-by-block bloom filtering for i := start; i <= end; i++ { blockNumber := rpc.BlockNumber(i) header, err := f.backend.HeaderByNumber(ctx, blockNumber) @@ -178,18 +256,12 @@ func (f *Filter) getLogs(ctx context.Context, start, end uint64) (logs []*types. // Use bloom filtering to see if this block is interesting given the // current parameters if f.bloomFilter(header.Bloom) { - // Get the logs of the block - receipts, err := f.backend.GetReceipts(ctx, header.Hash()) + logs, err := f.checkMatches(ctx, header) if err != nil { return nil, end, err } - var unfiltered []*types.Log - for _, receipt := range receipts { - unfiltered = append(unfiltered, ([]*types.Log)(receipt.Logs)...) - } - logs = filterLogs(unfiltered, nil, nil, f.addresses, f.topics) - if len(logs) > 0 { - return logs, uint64(blockNumber), nil + if logs != nil { + return logs, i, nil } } } diff --git a/eth/filters/filter_system_test.go b/eth/filters/filter_system_test.go index fcc888b8c..140fad555 100644 --- a/eth/filters/filter_system_test.go +++ b/eth/filters/filter_system_test.go @@ -36,6 +36,7 @@ import ( type testBackend struct { mux *event.TypeMux db ethdb.Database + sections uint64 txFeed *event.Feed rmLogsFeed *event.Feed logsFeed *event.Feed @@ -84,6 +85,31 @@ func (b *testBackend) SubscribeChainEvent(ch chan<- core.ChainEvent) event.Subsc return b.chainFeed.Subscribe(ch) } +func (b *testBackend) GetBloomBits(ctx context.Context, bitIdx uint64, sectionIdxList []uint64) ([][]byte, error) { + results := make([][]byte, len(sectionIdxList)) + var err error + for i, sectionIdx := range sectionIdxList { + sectionHead := core.GetCanonicalHash(b.db, (sectionIdx+1)*testBloomBitsSection-1) + results[i], err = core.GetBloomBits(b.db, bitIdx, sectionIdx, sectionHead) + if err != nil { + return nil, err + } + } + return results, nil +} + +func (b *testBackend) BloomBitsSections() uint64 { + return b.sections +} + +func (b *testBackend) BloomBitsConfig() BloomConfig { + return BloomConfig{ + SectionSize: testBloomBitsSection, + MaxRequestLen: 16, + MaxRequestWait: 0, + } +} + // TestBlockSubscription tests if a block subscription returns block hashes for posted chain events. // It creates multiple subscriptions: // - one at the start and should receive all posted chain events and a second (blockHashes) @@ -99,8 +125,8 @@ func TestBlockSubscription(t *testing.T) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} - api = NewPublicFilterAPI(backend, false) + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} + api = NewPublicFilterAPI(backend, false, 0) genesis = new(core.Genesis).MustCommit(db) chain, _ = core.GenerateChain(params.TestChainConfig, genesis, db, 10, func(i int, gen *core.BlockGen) {}) chainEvents = []core.ChainEvent{} @@ -156,8 +182,8 @@ func TestPendingTxFilter(t *testing.T) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} - api = NewPublicFilterAPI(backend, false) + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} + api = NewPublicFilterAPI(backend, false, 0) transactions = []*types.Transaction{ types.NewTransaction(0, common.HexToAddress("0xb794f5ea0ba39494ce83a213fffba74279579268"), new(big.Int), new(big.Int), new(big.Int), nil), @@ -219,8 +245,8 @@ func TestLogFilterCreation(t *testing.T) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} - api = NewPublicFilterAPI(backend, false) + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} + api = NewPublicFilterAPI(backend, false, 0) testCases = []struct { crit FilterCriteria @@ -268,8 +294,8 @@ func TestInvalidLogFilterCreation(t *testing.T) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} - api = NewPublicFilterAPI(backend, false) + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} + api = NewPublicFilterAPI(backend, false, 0) ) // different situations where log filter creation should fail. @@ -298,8 +324,8 @@ func TestLogFilter(t *testing.T) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} - api = NewPublicFilterAPI(backend, false) + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} + api = NewPublicFilterAPI(backend, false, 0) firstAddr = common.HexToAddress("0x1111111111111111111111111111111111111111") secondAddr = common.HexToAddress("0x2222222222222222222222222222222222222222") @@ -415,8 +441,8 @@ func TestPendingLogsSubscription(t *testing.T) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} - api = NewPublicFilterAPI(backend, false) + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} + api = NewPublicFilterAPI(backend, false, 0) firstAddr = common.HexToAddress("0x1111111111111111111111111111111111111111") secondAddr = common.HexToAddress("0x2222222222222222222222222222222222222222") diff --git a/eth/filters/filter_test.go b/eth/filters/filter_test.go index cf508a218..f1c6481d7 100644 --- a/eth/filters/filter_test.go +++ b/eth/filters/filter_test.go @@ -32,6 +32,8 @@ import ( "github.com/ethereum/go-ethereum/params" ) +const testBloomBitsSection = 4096 + func makeReceipt(addr common.Address) *types.Receipt { receipt := types.NewReceipt(nil, false, new(big.Int)) receipt.Logs = []*types.Log{ @@ -41,8 +43,8 @@ func makeReceipt(addr common.Address) *types.Receipt { return receipt } -func BenchmarkMipmaps(b *testing.B) { - dir, err := ioutil.TempDir("", "mipmap") +func BenchmarkFilters(b *testing.B) { + dir, err := ioutil.TempDir("", "filtertest") if err != nil { b.Fatal(err) } @@ -55,7 +57,7 @@ func BenchmarkMipmaps(b *testing.B) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} key1, _ = crypto.HexToECDSA("b71c71a67e1177ad4e901695e1b4b9ee17ae16c6668d313eac2f96dbcda3f291") addr1 = crypto.PubkeyToAddress(key1.PublicKey) addr2 = common.BytesToAddress([]byte("jeff")) @@ -66,27 +68,21 @@ func BenchmarkMipmaps(b *testing.B) { genesis := core.GenesisBlockForTesting(db, addr1, big.NewInt(1000000)) chain, receipts := core.GenerateChain(params.TestChainConfig, genesis, db, 100010, func(i int, gen *core.BlockGen) { - var receipts types.Receipts switch i { case 2403: receipt := makeReceipt(addr1) - receipts = types.Receipts{receipt} gen.AddUncheckedReceipt(receipt) case 1034: receipt := makeReceipt(addr2) - receipts = types.Receipts{receipt} gen.AddUncheckedReceipt(receipt) case 34: receipt := makeReceipt(addr3) - receipts = types.Receipts{receipt} gen.AddUncheckedReceipt(receipt) case 99999: receipt := makeReceipt(addr4) - receipts = types.Receipts{receipt} gen.AddUncheckedReceipt(receipt) } - core.WriteMipmapBloom(db, uint64(i+1), receipts) }) for i, block := range chain { core.WriteBlock(db, block) @@ -102,10 +98,7 @@ func BenchmarkMipmaps(b *testing.B) { } b.ResetTimer() - filter := New(backend, true) - filter.SetAddresses([]common.Address{addr1, addr2, addr3, addr4}) - filter.SetBeginBlock(0) - filter.SetEndBlock(-1) + filter := New(backend, 0, -1, []common.Address{addr1, addr2, addr3, addr4}, nil) for i := 0; i < b.N; i++ { logs, _ := filter.Find(context.Background()) @@ -116,7 +109,7 @@ func BenchmarkMipmaps(b *testing.B) { } func TestFilters(t *testing.T) { - dir, err := ioutil.TempDir("", "mipmap") + dir, err := ioutil.TempDir("", "filtertest") if err != nil { t.Fatal(err) } @@ -129,7 +122,7 @@ func TestFilters(t *testing.T) { rmLogsFeed = new(event.Feed) logsFeed = new(event.Feed) chainFeed = new(event.Feed) - backend = &testBackend{mux, db, txFeed, rmLogsFeed, logsFeed, chainFeed} + backend = &testBackend{mux, db, 0, txFeed, rmLogsFeed, logsFeed, chainFeed} key1, _ = crypto.HexToECDSA("b71c71a67e1177ad4e901695e1b4b9ee17ae16c6668d313eac2f96dbcda3f291") addr = crypto.PubkeyToAddress(key1.PublicKey) @@ -142,7 +135,6 @@ func TestFilters(t *testing.T) { genesis := core.GenesisBlockForTesting(db, addr, big.NewInt(1000000)) chain, receipts := core.GenerateChain(params.TestChainConfig, genesis, db, 1000, func(i int, gen *core.BlockGen) { - var receipts types.Receipts switch i { case 1: receipt := types.NewReceipt(nil, false, new(big.Int)) @@ -153,7 +145,6 @@ func TestFilters(t *testing.T) { }, } gen.AddUncheckedReceipt(receipt) - receipts = types.Receipts{receipt} case 2: receipt := types.NewReceipt(nil, false, new(big.Int)) receipt.Logs = []*types.Log{ @@ -163,7 +154,6 @@ func TestFilters(t *testing.T) { }, } gen.AddUncheckedReceipt(receipt) - receipts = types.Receipts{receipt} case 998: receipt := types.NewReceipt(nil, false, new(big.Int)) receipt.Logs = []*types.Log{ @@ -173,7 +163,6 @@ func TestFilters(t *testing.T) { }, } gen.AddUncheckedReceipt(receipt) - receipts = types.Receipts{receipt} case 999: receipt := types.NewReceipt(nil, false, new(big.Int)) receipt.Logs = []*types.Log{ @@ -183,12 +172,7 @@ func TestFilters(t *testing.T) { }, } gen.AddUncheckedReceipt(receipt) - receipts = types.Receipts{receipt} } - // i is used as block number for the writes but since the i - // starts at 0 and block 0 (genesis) is already present increment - // by one - core.WriteMipmapBloom(db, uint64(i+1), receipts) }) for i, block := range chain { core.WriteBlock(db, block) @@ -203,22 +187,14 @@ func TestFilters(t *testing.T) { } } - filter := New(backend, true) - filter.SetAddresses([]common.Address{addr}) - filter.SetTopics([][]common.Hash{{hash1, hash2, hash3, hash4}}) - filter.SetBeginBlock(0) - filter.SetEndBlock(-1) + filter := New(backend, 0, -1, []common.Address{addr}, [][]common.Hash{{hash1, hash2, hash3, hash4}}) logs, _ := filter.Find(context.Background()) if len(logs) != 4 { t.Error("expected 4 log, got", len(logs)) } - filter = New(backend, true) - filter.SetAddresses([]common.Address{addr}) - filter.SetTopics([][]common.Hash{{hash3}}) - filter.SetBeginBlock(900) - filter.SetEndBlock(999) + filter = New(backend, 900, 999, []common.Address{addr}, [][]common.Hash{{hash3}}) logs, _ = filter.Find(context.Background()) if len(logs) != 1 { t.Error("expected 1 log, got", len(logs)) @@ -227,11 +203,7 @@ func TestFilters(t *testing.T) { t.Errorf("expected log[0].Topics[0] to be %x, got %x", hash3, logs[0].Topics[0]) } - filter = New(backend, true) - filter.SetAddresses([]common.Address{addr}) - filter.SetTopics([][]common.Hash{{hash3}}) - filter.SetBeginBlock(990) - filter.SetEndBlock(-1) + filter = New(backend, 990, -1, []common.Address{addr}, [][]common.Hash{{hash3}}) logs, _ = filter.Find(context.Background()) if len(logs) != 1 { t.Error("expected 1 log, got", len(logs)) @@ -240,10 +212,7 @@ func TestFilters(t *testing.T) { t.Errorf("expected log[0].Topics[0] to be %x, got %x", hash3, logs[0].Topics[0]) } - filter = New(backend, true) - filter.SetTopics([][]common.Hash{{hash1, hash2}}) - filter.SetBeginBlock(1) - filter.SetEndBlock(10) + filter = New(backend, 1, 10, nil, [][]common.Hash{{hash1, hash2}}) logs, _ = filter.Find(context.Background()) if len(logs) != 2 { @@ -251,10 +220,7 @@ func TestFilters(t *testing.T) { } failHash := common.BytesToHash([]byte("fail")) - filter = New(backend, true) - filter.SetTopics([][]common.Hash{{failHash}}) - filter.SetBeginBlock(0) - filter.SetEndBlock(-1) + filter = New(backend, 0, -1, nil, [][]common.Hash{{failHash}}) logs, _ = filter.Find(context.Background()) if len(logs) != 0 { @@ -262,20 +228,14 @@ func TestFilters(t *testing.T) { } failAddr := common.BytesToAddress([]byte("failmenow")) - filter = New(backend, true) - filter.SetAddresses([]common.Address{failAddr}) - filter.SetBeginBlock(0) - filter.SetEndBlock(-1) + filter = New(backend, 0, -1, []common.Address{failAddr}, nil) logs, _ = filter.Find(context.Background()) if len(logs) != 0 { t.Error("expected 0 log, got", len(logs)) } - filter = New(backend, true) - filter.SetTopics([][]common.Hash{{failHash}, {hash1}}) - filter.SetBeginBlock(0) - filter.SetEndBlock(-1) + filter = New(backend, 0, -1, nil, [][]common.Hash{{failHash}, {hash1}}) logs, _ = filter.Find(context.Background()) if len(logs) != 0 { diff --git a/eth/handler.go b/eth/handler.go index 28ae208c0..aadd771df 100644 --- a/eth/handler.go +++ b/eth/handler.go @@ -49,6 +49,8 @@ const ( // txChanSize is the size of channel listening to TxPreEvent. // The number is referenced from the size of tx pool. txChanSize = 4096 + + bloomBitsSection = 4096 ) var ( @@ -92,6 +94,8 @@ type ProtocolManager struct { quitSync chan struct{} noMorePeers chan struct{} + lesServer LesServer + // wait group is used for graceful shutdowns during downloading // and processing wg sync.WaitGroup @@ -114,6 +118,7 @@ func NewProtocolManager(config *params.ChainConfig, mode downloader.SyncMode, ne txsyncCh: make(chan *txsync), quitSync: make(chan struct{}), } + // Figure out whether to allow fast sync or not if mode == downloader.FastSync && blockchain.CurrentBlock().NumberU64() > 0 { log.Warn("Blockchain not empty, fast sync disabled") diff --git a/les/api_backend.go b/les/api_backend.go index 1323e8864..c2ba27028 100644 --- a/les/api_backend.go +++ b/les/api_backend.go @@ -19,6 +19,7 @@ package les import ( "context" "math/big" + "time" "github.com/ethereum/go-ethereum/accounts" "github.com/ethereum/go-ethereum/common" @@ -28,6 +29,7 @@ import ( "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/core/vm" "github.com/ethereum/go-ethereum/eth/downloader" + "github.com/ethereum/go-ethereum/eth/filters" "github.com/ethereum/go-ethereum/eth/gasprice" "github.com/ethereum/go-ethereum/ethdb" "github.com/ethereum/go-ethereum/event" @@ -171,3 +173,19 @@ func (b *LesApiBackend) EventMux() *event.TypeMux { func (b *LesApiBackend) AccountManager() *accounts.Manager { return b.eth.accountManager } + +func (b *LesApiBackend) GetBloomBits(ctx context.Context, bitIdx uint64, sectionIdxList []uint64) ([][]byte, error) { + return nil, nil // implemented in a subsequent PR +} + +func (b *LesApiBackend) BloomBitsSections() uint64 { + return 0 +} + +func (b *LesApiBackend) BloomBitsConfig() filters.BloomConfig { + return filters.BloomConfig{ + SectionSize: 32768, + MaxRequestLen: 16, + MaxRequestWait: time.Microsecond * 100, + } +} diff --git a/les/backend.go b/les/backend.go index 4c33417c0..a3670b5ac 100644 --- a/les/backend.go +++ b/les/backend.go @@ -169,7 +169,7 @@ func (s *LightEthereum) APIs() []rpc.API { }, { Namespace: "eth", Version: "1.0", - Service: filters.NewPublicFilterAPI(s.ApiBackend, true), + Service: filters.NewPublicFilterAPI(s.ApiBackend, true, 0), Public: true, }, { Namespace: "net", diff --git a/miner/worker.go b/miner/worker.go index 5bac5d6e8..e1154ac06 100644 --- a/miner/worker.go +++ b/miner/worker.go @@ -324,8 +324,6 @@ func (self *worker) wait() { if stat == core.CanonStatTy { // This puts transactions in a extra db for rpc core.WriteTxLookupEntries(self.chainDb, block) - // Write map map bloom filters - core.WriteMipmapBloom(self.chainDb, block.NumberU64(), work.receipts) // implicit by posting ChainHeadEvent mustCommitNewWork = false }