path: root/core/chain_indexer.go

// 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 (
	"context"
	"encoding/binary"
	"fmt"
	"sync"
	"sync/atomic"
	"time"

	"github.com/ava-labs/coreth/core/rawdb"
	"github.com/ava-labs/coreth/core/types"
	"github.com/ethereum/go-ethereum/common"
	"github.com/ethereum/go-ethereum/ethdb"
	"github.com/ethereum/go-ethereum/event"
	"github.com/ethereum/go-ethereum/log"
)

// 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(ctx context.Context, section uint64, prevHead common.Hash) error

	// Process crunches through the next header in the chain segment. The caller
	// will ensure a sequential order of headers.
	Process(ctx context.Context, header *types.Header) error

	// Commit finalizes the section metadata and stores it into the database.
	Commit() error

	// Prune deletes the chain index older than the given threshold.
	Prune(threshold uint64) error
}

// ChainIndexerChain interface is used for connecting the indexer to a blockchain
type ChainIndexerChain interface {
	// CurrentHeader retrieves the latest locally known header.
	CurrentHeader() *types.Header

	// SubscribeChainHeadEvent subscribes to new head header notifications.
	SubscribeChainHeadEvent(ch chan<- ChainHeadEvent) event.Subscription
}

// 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
// ChainHeadEventLoop 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
	ctx       context.Context
	ctxCancel func()

	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

	checkpointSections uint64      // Number of sections covered by the checkpoint
	checkpointHead     common.Hash // Section head belonging to the checkpoint

	throttling time.Duration // Disk throttling to prevent a heavy upgrade from hogging resources

	log  log.Logger
	lock sync.RWMutex
}

// 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 ethdb.Database, indexDb ethdb.Database, backend ChainIndexerBackend, section, confirm uint64, throttling time.Duration, kind string) *ChainIndexer {
	c := &ChainIndexer{
		chainDb:     chainDb,
		indexDb:     indexDb,
		backend:     backend,
		update:      make(chan struct{}, 1),
		quit:        make(chan chan error),
		sectionSize: section,
		confirmsReq: confirm,
		throttling:  throttling,
		log:         log.New("type", kind),
	}
	// Initialize database dependent fields and start the updater
	c.loadValidSections()
	c.ctx, c.ctxCancel = context.WithCancel(context.Background())

	go c.updateLoop()

	return c
}

// AddCheckpoint adds a checkpoint. Sections are never processed and the chain
// is not expected to be available before this point. The indexer assumes that
// the backend has sufficient information available to process subsequent sections.
//
// Note: knownSections == 0 and storedSections == checkpointSections until
// syncing reaches the checkpoint
func (c *ChainIndexer) AddCheckpoint(section uint64, shead common.Hash) {
	c.lock.Lock()
	defer c.lock.Unlock()

	// Short circuit if the given checkpoint is below than local's.
	if c.checkpointSections >= section+1 || section < c.storedSections {
		return
	}
	c.checkpointSections = section + 1
	c.checkpointHead = shead

	c.setSectionHead(section, shead)
	c.setValidSections(section + 1)
}

// 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 ChainHeadEvent, 10)
	sub := chain.SubscribeChainHeadEvent(events)

	go c.eventLoop(chain.CurrentHeader(), events, sub)
}

// 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

	c.ctxCancel()

	// 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)
		}
	}
	// Return any failures
	switch {
	case len(errs) == 0:
		return nil

	case len(errs) == 1:
		return errs[0]

	default:
		return fmt.Errorf("%v", errs)
	}
}

// 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 ChainHeadEvent, sub event.Subscription) {
	// 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
	c.newHead(currentHeader.Number.Uint64(), false)

	var (
		prevHeader = currentHeader
		prevHash   = currentHeader.Hash()
	)
	for {
		select {
		case errc := <-c.quit:
			// Chain indexer terminating, report no failure and abort
			errc <- nil
			return

		case ev, ok := <-events:
			// Received a new event, ensure it's not nil (closing) and update
			if !ok {
				errc := <-c.quit
				errc <- nil
				return
			}
			header := ev.Block.Header()
			if header.ParentHash != prevHash {
				// Reorg to the common ancestor if needed (might not exist in light sync mode, skip reorg then)
				// TODO(karalabe, zsfelfoldi): This seems a bit brittle, can we detect this case explicitly?

				if rawdb.ReadCanonicalHash(c.chainDb, prevHeader.Number.Uint64()) != prevHash {
					if h := rawdb.FindCommonAncestor(c.chainDb, prevHeader, header); h != nil {
						c.newHead(h.Number.Uint64(), true)
					}
				}
			}
			c.newHead(header.Number.Uint64(), false)

			prevHeader, prevHash = header, header.Hash()
		}
	}
}

// newHead notifies the indexer about new chain heads and/or reorgs.
func (c *ChainIndexer) newHead(head uint64, reorg bool) {
	c.lock.Lock()
	defer c.lock.Unlock()

	// If a reorg happened, invalidate all sections until that point
	if reorg {
		// Revert the known section number to the reorg point
		known := (head + 1) / c.sectionSize
		stored := known
		if known < c.checkpointSections {
			known = 0
		}
		if stored < c.checkpointSections {
			stored = c.checkpointSections
		}
		if known < c.knownSections {
			c.knownSections = known
		}
		// Revert the stored sections from the database to the reorg point
		if stored < c.storedSections {
			c.setValidSections(stored)
		}
		// Update the new head number to the finalized section end and notify children
		head = known * c.sectionSize

		if head < c.cascadedHead {
			c.cascadedHead = head
			for _, child := range c.children {
				child