plugeth/core/chain_indexer.go

449 lines
15 KiB
Go
Raw Normal View History

2017-03-05 15:52:03 +00:00
// 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 <http://www.gnu.org/licenses/>.
package core
import (
"encoding/binary"
2017-08-03 16:25:06 +00:00
"fmt"
2017-03-05 15:52:03 +00:00
"sync"
2017-08-03 16:25:06 +00:00
"sync/atomic"
2017-03-05 15:52:03 +00:00
"time"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/event"
2017-08-03 16:25:06 +00:00
"github.com/ethereum/go-ethereum/log"
2017-03-05 15:52:03 +00:00
)
2017-08-03 16:25:06 +00:00
// ChainIndexerBackend defines the methods needed to process chain segments in
// the background and write the segment results into the database. These can be
// used to create filter blooms or CHTs.
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, prevHead common.Hash) error
2017-08-03 16:25:06 +00:00
// Process crunches through the next header in the chain segment. The caller
// will ensure a sequential order of headers.
Process(header *types.Header)
// Commit finalizes the section metadata and stores it into the database.
Commit() error
2017-03-05 15:52:03 +00:00
}
// ChainIndexerChain interface is used for connecting the indexer to a blockchain
type ChainIndexerChain interface {
// CurrentHeader retrieves the latest locally known header.
CurrentHeader() *types.Header
// SubscribeChainEvent subscribes to new head header notifications.
SubscribeChainEvent(ch chan<- ChainEvent) event.Subscription
}
2017-08-03 16:25:06 +00:00
// ChainIndexer does a post-processing job for equally sized sections of the
// canonical chain (like BlooomBits and CHT structures). A ChainIndexer is
// connected to the blockchain through the event system by starting a
// ChainEventLoop in a goroutine.
//
// Further child ChainIndexers can be added which use the output of the parent
// section indexer. These child indexers receive new head notifications only
// after an entire section has been finished or in case of rollbacks that might
// affect already finished sections.
type ChainIndexer struct {
chainDb ethdb.Database // Chain database to index the data from
indexDb ethdb.Database // Prefixed table-view of the db to write index metadata into
backend ChainIndexerBackend // Background processor generating the index data content
children []*ChainIndexer // Child indexers to cascade chain updates to
active uint32 // Flag whether the event loop was started
update chan struct{} // Notification channel that headers should be processed
quit chan chan error // Quit channel to tear down running goroutines
sectionSize uint64 // Number of blocks in a single chain segment to process
confirmsReq uint64 // Number of confirmations before processing a completed segment
storedSections uint64 // Number of sections successfully indexed into the database
knownSections uint64 // Number of sections known to be complete (block wise)
cascadedHead uint64 // Block number of the last completed section cascaded to subindexers
throttling time.Duration // Disk throttling to prevent a heavy upgrade from hogging resources
log log.Logger
lock sync.RWMutex
2017-03-05 15:52:03 +00:00
}
2017-08-03 16:25:06 +00:00
// NewChainIndexer creates a new chain indexer to do background processing on
// chain segments of a given size after certain number of confirmations passed.
// The throttling parameter might be used to prevent database thrashing.
func NewChainIndexer(chainDb, indexDb ethdb.Database, backend ChainIndexerBackend, section, confirm uint64, throttling time.Duration, kind string) *ChainIndexer {
2017-03-05 15:52:03 +00:00
c := &ChainIndexer{
chainDb: chainDb,
indexDb: indexDb,
backend: backend,
2017-08-03 16:25:06 +00:00
update: make(chan struct{}, 1),
quit: make(chan chan error),
sectionSize: section,
confirmsReq: confirm,
throttling: throttling,
log: log.New("type", kind),
2017-03-05 15:52:03 +00:00
}
2017-08-03 16:25:06 +00:00
// Initialize database dependent fields and start the updater
c.loadValidSections()
2017-03-05 15:52:03 +00:00
go c.updateLoop()
2017-08-03 16:25:06 +00:00
2017-03-05 15:52:03 +00:00
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)
}
2017-08-03 16:25:06 +00:00
// Start creates a goroutine to feed chain head events into the indexer for
// cascading background processing. Children do not need to be started, they
// are notified about new events by their parents.
func (c *ChainIndexer) Start(chain ChainIndexerChain) {
events := make(chan ChainEvent, 10)
sub := chain.SubscribeChainEvent(events)
go c.eventLoop(chain.CurrentHeader(), events, sub)
2017-08-03 16:25:06 +00:00
}
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
// Close tears down all goroutines belonging to the indexer and returns any error
// that might have occurred internally.
func (c *ChainIndexer) Close() error {
var errs []error
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
// Tear down the primary update loop
errc := make(chan error)
c.quit <- errc
if err := <-errc; err != nil {
errs = append(errs, err)
}
// If needed, tear down the secondary event loop
if atomic.LoadUint32(&c.active) != 0 {
c.quit <- errc
if err := <-errc; err != nil {
errs = append(errs, err)
}
}
// Close all children
for _, child := range c.children {
if err := child.Close(); err != nil {
errs = append(errs, err)
}
}
2017-08-03 16:25:06 +00:00
// Return any failures
switch {
case len(errs) == 0:
return nil
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
case len(errs) == 1:
return errs[0]
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
default:
return fmt.Errorf("%v", errs)
2017-03-05 15:52:03 +00:00
}
}
2017-08-03 16:25:06 +00:00
// 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, events chan ChainEvent, sub event.Subscription) {
2017-08-03 16:25:06 +00:00
// Mark the chain indexer as active, requiring an additional teardown
atomic.StoreUint32(&c.active, 1)
defer sub.Unsubscribe()
// Fire the initial new head event to start any outstanding processing
2017-03-05 15:52:03 +00:00
c.newHead(currentHeader.Number.Uint64(), false)
2017-08-03 16:25:06 +00:00
var (
prevHeader = currentHeader
prevHash = currentHeader.Hash()
)
2017-03-05 15:52:03 +00:00
for {
select {
2017-08-03 16:25:06 +00:00
case errc := <-c.quit:
// Chain indexer terminating, report no failure and abort
errc <- nil
2017-03-05 15:52:03 +00:00
return
2017-08-03 16:25:06 +00:00
case ev, ok := <-events:
2017-08-03 16:25:06 +00:00
// Received a new event, ensure it's not nil (closing) and update
if !ok {
errc := <-c.quit
errc <- nil
return
}
header := ev.Block.Header()
2017-08-03 16:25:06 +00:00
if header.ParentHash != prevHash {
// Reorg to the common ancestor (might not exist in light sync mode, skip reorg then)
// TODO(karalabe, zsfelfoldi): This seems a bit brittle, can we detect this case explicitly?
// TODO(karalabe): This operation is expensive and might block, causing the event system to
// potentially also lock up. We need to do with on a different thread somehow.
if h := FindCommonAncestor(c.chainDb, prevHeader, header); h != nil {
c.newHead(h.Number.Uint64(), true)
}
2017-08-03 16:25:06 +00:00
}
c.newHead(header.Number.Uint64(), false)
prevHeader, prevHash = header, header.Hash()
2017-03-05 15:52:03 +00:00
}
}
}
2017-08-03 16:25:06 +00:00
// newHead notifies the indexer about new chain heads and/or reorgs.
func (c *ChainIndexer) newHead(head uint64, reorg bool) {
2017-03-05 15:52:03 +00:00
c.lock.Lock()
defer c.lock.Unlock()
2017-08-03 16:25:06 +00:00
// If a reorg happened, invalidate all sections until that point
if reorg {
// Revert the known section number to the reorg point
changed := head / c.sectionSize
if changed < c.knownSections {
c.knownSections = changed
2017-03-05 15:52:03 +00:00
}
2017-08-03 16:25:06 +00:00
// Revert the stored sections from the database to the reorg point
if changed < c.storedSections {
c.setValidSections(changed)
2017-03-05 15:52:03 +00:00
}
// Update the new head number to the finalized section end and notify children
2017-08-03 16:25:06 +00:00
head = changed * c.sectionSize
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
if head < c.cascadedHead {
c.cascadedHead = head
for _, child := range c.children {
child.newHead(c.cascadedHead, true)
2017-03-05 15:52:03 +00:00
}
}
2017-08-03 16:25:06 +00:00
return
}
// No reorg, calculate the number of newly known sections and update if high enough
var sections uint64
if head >= c.confirmsReq {
sections = (head + 1 - c.confirmsReq) / c.sectionSize
if sections > c.knownSections {
c.knownSections = sections
select {
case c.update <- struct{}{}:
default:
}
}
}
}
// updateLoop is the main event loop of the indexer which pushes chain segments
// down into the processing backend.
func (c *ChainIndexer) updateLoop() {
var (
updating bool
updated time.Time
)
2017-08-03 16:25:06 +00:00
for {
select {
case errc := <-c.quit:
// Chain indexer terminating, report no failure and abort
errc <- nil
return
case <-c.update:
// Section headers completed (or rolled back), update the index
c.lock.Lock()
if c.knownSections > c.storedSections {
// Periodically print an upgrade log message to the user
if time.Since(updated) > 8*time.Second {
if c.knownSections > c.storedSections+1 {
updating = true
2017-08-03 16:25:06 +00:00
c.log.Info("Upgrading chain index", "percentage", c.storedSections*100/c.knownSections)
}
updated = time.Now()
}
// Cache the current section count and head to allow unlocking the mutex
section := c.storedSections
var oldHead common.Hash
if section > 0 {
oldHead = c.SectionHead(section - 1)
2017-08-03 16:25:06 +00:00
}
// 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)
}
2017-08-03 16:25:06 +00:00
c.lock.Lock()
// If processing succeeded and no reorgs occcurred, mark the section completed
if err == nil && oldHead == c.SectionHead(section-1) {
2017-08-03 16:25:06 +00:00
c.setSectionHead(section, newHead)
c.setValidSections(section + 1)
if c.storedSections == c.knownSections && updating {
updating = false
c.log.Info("Finished upgrading chain index")
}
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
c.cascadedHead = c.storedSections*c.sectionSize - 1
for _, child := range c.children {
c.log.Trace("Cascading chain index update", "head", c.cascadedHead)
child.newHead(c.cascadedHead, false)
}
} else {
// If processing failed, don't retry until further notification
c.log.Debug("Chain index processing failed", "section", section, "err", err)
c.knownSections = c.storedSections
2017-03-05 15:52:03 +00:00
}
}
2017-08-03 16:25:06 +00:00
// If there are still further sections to process, reschedule
if c.knownSections > c.storedSections {
time.AfterFunc(c.throttling, func() {
select {
case c.update <- struct{}{}:
default:
}
})
}
c.lock.Unlock()
2017-03-05 15:52:03 +00:00
}
}
}
2017-08-03 16:25:06 +00:00
// processSection processes an entire section by calling backend functions while
// ensuring the continuity of the passed headers. Since the chain mutex is not
// held while processing, the continuity can be broken by a long reorg, in which
// case the function returns with an error.
func (c *ChainIndexer) processSection(section uint64, lastHead common.Hash) (common.Hash, error) {
c.log.Trace("Processing new chain section", "section", section)
// Reset and partial processing
if err := c.backend.Reset(section, lastHead); err != nil {
c.setValidSections(0)
return common.Hash{}, err
}
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
for number := section * c.sectionSize; number < (section+1)*c.sectionSize; number++ {
hash := GetCanonicalHash(c.chainDb, number)
2017-03-05 15:52:03 +00:00
if hash == (common.Hash{}) {
2017-08-03 16:25:06 +00:00
return common.Hash{}, fmt.Errorf("canonical block #%d unknown", number)
2017-03-05 15:52:03 +00:00
}
2017-08-03 16:25:06 +00:00
header := GetHeader(c.chainDb, hash, number)
if header == nil {
return common.Hash{}, fmt.Errorf("block #%d [%x…] not found", number, hash[:4])
} else if header.ParentHash != lastHead {
return common.Hash{}, fmt.Errorf("chain reorged during section processing")
2017-03-05 15:52:03 +00:00
}
c.backend.Process(header)
2017-08-03 16:25:06 +00:00
lastHead = header.Hash()
2017-03-05 15:52:03 +00:00
}
if err := c.backend.Commit(); err != nil {
c.log.Error("Section commit failed", "error", err)
2017-08-03 16:25:06 +00:00
return common.Hash{}, err
2017-03-05 15:52:03 +00:00
}
2017-08-03 16:25:06 +00:00
return lastHead, nil
2017-03-05 15:52:03 +00:00
}
2017-08-03 16:25:06 +00:00
// Sections returns the number of processed sections maintained by the indexer
// and also the information about the last header indexed for potential canonical
// verifications.
func (c *ChainIndexer) Sections() (uint64, uint64, common.Hash) {
2017-03-05 15:52:03 +00:00
c.lock.Lock()
defer c.lock.Unlock()
return c.storedSections, c.storedSections*c.sectionSize - 1, c.SectionHead(c.storedSections - 1)
2017-08-03 16:25:06 +00:00
}
// AddChildIndexer adds a child ChainIndexer that can use the output of this one
func (c *ChainIndexer) AddChildIndexer(indexer *ChainIndexer) {
c.lock.Lock()
defer c.lock.Unlock()
c.children = append(c.children, indexer)
// Cascade any pending updates to new children too
if c.storedSections > 0 {
indexer.newHead(c.storedSections*c.sectionSize-1, false)
2017-03-05 15:52:03 +00:00
}
}
2017-08-03 16:25:06 +00:00
// loadValidSections reads the number of valid sections from the index database
// and caches is into the local state.
func (c *ChainIndexer) loadValidSections() {
2017-03-05 15:52:03 +00:00
data, _ := c.indexDb.Get([]byte("count"))
if len(data) == 8 {
2017-08-03 16:25:06 +00:00
c.storedSections = binary.BigEndian.Uint64(data[:])
2017-03-05 15:52:03 +00:00
}
}
// setValidSections writes the number of valid sections to the index database
2017-08-03 16:25:06 +00:00
func (c *ChainIndexer) setValidSections(sections uint64) {
// Set the current number of valid sections in the database
2017-03-05 15:52:03 +00:00
var data [8]byte
2017-08-03 16:25:06 +00:00
binary.BigEndian.PutUint64(data[:], sections)
2017-03-05 15:52:03 +00:00
c.indexDb.Put([]byte("count"), data[:])
2017-08-03 16:25:06 +00:00
// Remove any reorged sections, caching the valids in the mean time
for c.storedSections > sections {
c.storedSections--
c.removeSectionHead(c.storedSections)
}
c.storedSections = sections // needed if new > old
2017-03-05 15:52:03 +00:00
}
// SectionHead retrieves the last block hash of a processed section from the
2017-08-03 16:25:06 +00:00
// index database.
func (c *ChainIndexer) SectionHead(section uint64) common.Hash {
2017-03-05 15:52:03 +00:00
var data [8]byte
2017-08-03 16:25:06 +00:00
binary.BigEndian.PutUint64(data[:], section)
2017-03-05 15:52:03 +00:00
hash, _ := c.indexDb.Get(append([]byte("shead"), data[:]...))
if len(hash) == len(common.Hash{}) {
return common.BytesToHash(hash)
}
return common.Hash{}
}
2017-08-03 16:25:06 +00:00
// setSectionHead writes the last block hash of a processed section to the index
// database.
func (c *ChainIndexer) setSectionHead(section uint64, hash common.Hash) {
2017-03-05 15:52:03 +00:00
var data [8]byte
2017-08-03 16:25:06 +00:00
binary.BigEndian.PutUint64(data[:], section)
2017-03-05 15:52:03 +00:00
2017-08-03 16:25:06 +00:00
c.indexDb.Put(append([]byte("shead"), data[:]...), hash.Bytes())
2017-03-05 15:52:03 +00:00
}
2017-08-03 16:25:06 +00:00
// removeSectionHead removes the reference to a processed section from the index
// database.
func (c *ChainIndexer) removeSectionHead(section uint64) {
2017-03-05 15:52:03 +00:00
var data [8]byte
2017-08-03 16:25:06 +00:00
binary.BigEndian.PutUint64(data[:], section)
2017-03-05 15:52:03 +00:00
c.indexDb.Delete(append([]byte("shead"), data[:]...))
}