prysm-pulse/beacon-chain/sync/pending_blocks_queue.go
terencechain c5501f8775
Refactor Pending Block Queue Logic in Sync Package (#13026)
* Clean up pending block queue

* Kasey's feedback

* Kasey's feedback on validateBeaconBlock err handling

* Clean up handleBlockProcessingError

* Clean up old comments

* James feedback

* has peer helper

* Reuse parent reoot

---------

Co-authored-by: prylabs-bulldozer[bot] <58059840+prylabs-bulldozer[bot]@users.noreply.github.com>
2023-10-11 14:41:24 +00:00

493 lines
14 KiB
Go

package sync
import (
"context"
"encoding/hex"
"fmt"
"sort"
"sync"
"time"
"github.com/pkg/errors"
"github.com/prysmaticlabs/prysm/v4/async"
"github.com/prysmaticlabs/prysm/v4/beacon-chain/blockchain"
p2ptypes "github.com/prysmaticlabs/prysm/v4/beacon-chain/p2p/types"
"github.com/prysmaticlabs/prysm/v4/config/params"
"github.com/prysmaticlabs/prysm/v4/consensus-types/blocks"
"github.com/prysmaticlabs/prysm/v4/consensus-types/interfaces"
"github.com/prysmaticlabs/prysm/v4/consensus-types/primitives"
"github.com/prysmaticlabs/prysm/v4/crypto/rand"
"github.com/prysmaticlabs/prysm/v4/encoding/bytesutil"
"github.com/prysmaticlabs/prysm/v4/encoding/ssz/equality"
"github.com/prysmaticlabs/prysm/v4/monitoring/tracing"
"github.com/prysmaticlabs/prysm/v4/time/slots"
"github.com/sirupsen/logrus"
"github.com/trailofbits/go-mutexasserts"
"go.opencensus.io/trace"
)
var processPendingBlocksPeriod = slots.DivideSlotBy(3 /* times per slot */)
const maxPeerRequest = 50
const numOfTries = 5
const maxBlocksPerSlot = 3
// processes pending blocks queue on every processPendingBlocksPeriod
func (s *Service) processPendingBlocksQueue() {
// Prevents multiple queue processing goroutines (invoked by RunEvery) from contending for data.
locker := new(sync.Mutex)
async.RunEvery(s.ctx, processPendingBlocksPeriod, func() {
// Don't process the pending blocks if genesis time has not been set. The chain is not ready.
if !s.chainIsStarted() {
return
}
locker.Lock()
if err := s.processPendingBlocks(s.ctx); err != nil {
log.WithError(err).Debug("Could not process pending blocks")
}
locker.Unlock()
})
}
// processPendingBlocks validates, processes, and broadcasts pending blocks.
func (s *Service) processPendingBlocks(ctx context.Context) error {
ctx, span := trace.StartSpan(ctx, "processPendingBlocks")
defer span.End()
// Validate pending slots before processing.
if err := s.validatePendingSlots(); err != nil {
return errors.Wrap(err, "could not validate pending slots")
}
// Sort slots for ordered processing.
sortedSlots := s.sortedPendingSlots()
span.AddAttributes(trace.Int64Attribute("numSlots", int64(len(sortedSlots))), trace.Int64Attribute("numPeers", int64(len(s.cfg.p2p.Peers().Connected()))))
randGen := rand.NewGenerator()
var parentRoots [][32]byte
// Iterate through sorted slots.
for _, slot := range sortedSlots {
// Skip processing if slot is in the future.
if slot > s.cfg.clock.CurrentSlot() {
continue
}
ctx, span := startInnerSpan(ctx, slot)
// Get blocks in the pending queue for the current slot.
blocksInCache := s.getBlocksInQueue(slot)
if len(blocksInCache) == 0 {
span.End()
continue
}
// Process each block in the queue.
for _, b := range blocksInCache {
if err := blocks.BeaconBlockIsNil(b); err != nil {
continue
}
blkRoot, err := b.Block().HashTreeRoot()
if err != nil {
return err
}
// Skip blocks that are already being processed.
if s.cfg.chain.BlockBeingSynced(blkRoot) {
log.WithField("BlockRoot", fmt.Sprintf("%#x", blkRoot)).Info("Skipping pending block already being processed")
continue
}
// Remove and skip blocks already in the database.
if s.cfg.beaconDB.HasBlock(ctx, blkRoot) {
if err := s.removeBlockFromQueue(b, blkRoot); err != nil {
return err
}
continue
}
parentRoot := b.Block().ParentRoot()
inPendingQueue := s.isBlockInQueue(parentRoot)
// Check if block is bad.
keepProcessing, err := s.checkIfBlockIsBad(ctx, span, slot, b, blkRoot)
if err != nil {
return err
}
if !keepProcessing {
continue
}
// Request parent block if not in the pending queue and not in the database.
isParentBlockInDB := s.cfg.beaconDB.HasBlock(ctx, parentRoot)
if !inPendingQueue && !isParentBlockInDB && s.hasPeer() {
log.WithFields(logrus.Fields{"currentSlot": b.Block().Slot(), "parentRoot": hex.EncodeToString(parentRoot[:])}).Debug("Requesting parent block")
parentRoots = append(parentRoots, parentRoot)
continue
}
if !isParentBlockInDB {
continue
}
// Process and broadcast the block.
if err := s.processAndBroadcastBlock(ctx, b, blkRoot); err != nil {
s.handleBlockProcessingError(ctx, err, b, blkRoot)
continue
}
// Remove the processed block from the queue.
if err := s.removeBlockFromQueue(b, blkRoot); err != nil {
return err
}
log.WithFields(logrus.Fields{"slot": slot, "blockRoot": hex.EncodeToString(bytesutil.Trunc(blkRoot[:]))}).Debug("Processed pending block and cleared it in cache")
}
span.End()
}
return s.sendBatchRootRequest(ctx, parentRoots, randGen)
}
// startInnerSpan starts a new tracing span for an inner loop and returns the new context and span.
func startInnerSpan(ctx context.Context, slot primitives.Slot) (context.Context, *trace.Span) {
ctx, span := trace.StartSpan(ctx, "processPendingBlocks.InnerLoop")
span.AddAttributes(trace.Int64Attribute("slot", int64(slot))) // lint:ignore uintcast -- This conversion is OK for tracing.
return ctx, span
}
// getBlocksInQueue retrieves the blocks in the pending queue for a given slot.
func (s *Service) getBlocksInQueue(slot primitives.Slot) []interfaces.ReadOnlySignedBeaconBlock {
s.pendingQueueLock.RLock()
defer s.pendingQueueLock.RUnlock()
return s.pendingBlocksInCache(slot)
}
// removeBlockFromQueue removes a block from the pending queue.
func (s *Service) removeBlockFromQueue(b interfaces.ReadOnlySignedBeaconBlock, blkRoot [32]byte) error {
s.pendingQueueLock.Lock()
defer s.pendingQueueLock.Unlock()
if err := s.deleteBlockFromPendingQueue(b.Block().Slot(), b, blkRoot); err != nil {
return err
}
return nil
}
// isBlockInQueue checks if a block's parent root is in the pending queue.
func (s *Service) isBlockInQueue(parentRoot [32]byte) bool {
s.pendingQueueLock.RLock()
defer s.pendingQueueLock.RUnlock()
return s.seenPendingBlocks[parentRoot]
}
func (s *Service) hasPeer() bool {
return len(s.cfg.p2p.Peers().Connected()) > 0
}
// processAndBroadcastBlock validates, processes, and broadcasts a block.
func (s *Service) processAndBroadcastBlock(ctx context.Context, b interfaces.ReadOnlySignedBeaconBlock, blkRoot [32]byte) error {
if err := s.validateBeaconBlock(ctx, b, blkRoot); err != nil {
if !errors.Is(ErrOptimisticParent, err) {
log.WithError(err).WithField("slot", b.Block().Slot()).Debug("Could not validate block")
return err
}
}
if err := s.cfg.chain.ReceiveBlock(ctx, b, blkRoot); err != nil {
return err
}
s.setSeenBlockIndexSlot(b.Block().Slot(), b.Block().ProposerIndex())
pb, err := b.Proto()
if err != nil {
log.WithError(err).Debug("Could not get protobuf block")
return err
}
if err := s.cfg.p2p.Broadcast(ctx, pb); err != nil {
log.WithError(err).Debug("Could not broadcast block")
return err
}
return nil
}
// handleBlockProcessingError handles errors during block processing.
func (s *Service) handleBlockProcessingError(ctx context.Context, err error, b interfaces.ReadOnlySignedBeaconBlock, blkRoot [32]byte) {
if blockchain.IsInvalidBlock(err) {
s.setBadBlock(ctx, blkRoot)
}
log.WithError(err).WithField("slot", b.Block().Slot()).Debug("Could not process block")
}
func (s *Service) checkIfBlockIsBad(
ctx context.Context,
span *trace.Span,
slot primitives.Slot,
b interfaces.ReadOnlySignedBeaconBlock,
blkRoot [32]byte,
) (keepProcessing bool, err error) {
parentIsBad := s.hasBadBlock(b.Block().ParentRoot())
blockIsBad := s.hasBadBlock(blkRoot)
// Check if parent is a bad block.
if parentIsBad || blockIsBad {
// Set block as bad if its parent block is bad too.
if parentIsBad {
s.setBadBlock(ctx, blkRoot)
}
// Remove block from queue.
s.pendingQueueLock.Lock()
if err = s.deleteBlockFromPendingQueue(slot, b, blkRoot); err != nil {
s.pendingQueueLock.Unlock()
return false, err
}
s.pendingQueueLock.Unlock()
span.End()
return false, nil
}
return true, nil
}
func (s *Service) sendBatchRootRequest(ctx context.Context, roots [][32]byte, randGen *rand.Rand) error {
ctx, span := trace.StartSpan(ctx, "sendBatchRootRequest")
defer span.End()
roots = dedupRoots(roots)
s.pendingQueueLock.RLock()
for i := len(roots) - 1; i >= 0; i-- {
r := roots[i]
if s.seenPendingBlocks[r] || s.cfg.chain.BlockBeingSynced(r) {
roots = append(roots[:i], roots[i+1:]...)
}
}
s.pendingQueueLock.RUnlock()
if len(roots) == 0 {
return nil
}
cp := s.cfg.chain.FinalizedCheckpt()
_, bestPeers := s.cfg.p2p.Peers().BestFinalized(maxPeerRequest, cp.Epoch)
if len(bestPeers) == 0 {
return nil
}
// Randomly choose a peer to query from our best peers. If that peer cannot return
// all the requested blocks, we randomly select another peer.
pid := bestPeers[randGen.Int()%len(bestPeers)]
for i := 0; i < numOfTries; i++ {
req := p2ptypes.BeaconBlockByRootsReq(roots)
if len(roots) > int(params.BeaconNetworkConfig().MaxRequestBlocks) {
req = roots[:params.BeaconNetworkConfig().MaxRequestBlocks]
}
if err := s.sendRecentBeaconBlocksRequest(ctx, &req, pid); err != nil {
tracing.AnnotateError(span, err)
log.WithError(err).Debug("Could not send recent block request")
}
newRoots := make([][32]byte, 0, len(roots))
s.pendingQueueLock.RLock()
for _, rt := range roots {
if !s.seenPendingBlocks[rt] {
newRoots = append(newRoots, rt)
}
}
s.pendingQueueLock.RUnlock()
if len(newRoots) == 0 {
break
}
// Choosing a new peer with the leftover set of
// roots to request.
roots = newRoots
pid = bestPeers[randGen.Int()%len(bestPeers)]
}
return nil
}
func (s *Service) sortedPendingSlots() []primitives.Slot {
s.pendingQueueLock.RLock()
defer s.pendingQueueLock.RUnlock()
items := s.slotToPendingBlocks.Items()
ss := make([]primitives.Slot, 0, len(items))
for k := range items {
slot := cacheKeyToSlot(k)
ss = append(ss, slot)
}
sort.Slice(ss, func(i, j int) bool {
return ss[i] < ss[j]
})
return ss
}
// validatePendingSlots validates the pending blocks
// by their slot. If they are before the current finalized
// checkpoint, these blocks are removed from the queue.
func (s *Service) validatePendingSlots() error {
s.pendingQueueLock.Lock()
defer s.pendingQueueLock.Unlock()
oldBlockRoots := make(map[[32]byte]bool)
cp := s.cfg.chain.FinalizedCheckpt()
finalizedEpoch := cp.Epoch
if s.slotToPendingBlocks == nil {
return errors.New("slotToPendingBlocks cache can't be nil")
}
items := s.slotToPendingBlocks.Items()
for k := range items {
slot := cacheKeyToSlot(k)
blks := s.pendingBlocksInCache(slot)
for _, b := range blks {
epoch := slots.ToEpoch(slot)
// remove all descendant blocks of old blocks
if oldBlockRoots[b.Block().ParentRoot()] {
root, err := b.Block().HashTreeRoot()
if err != nil {
return err
}
oldBlockRoots[root] = true
if err := s.deleteBlockFromPendingQueue(slot, b, root); err != nil {
return err
}
continue
}
// don't process old blocks
if finalizedEpoch > 0 && epoch <= finalizedEpoch {
blkRoot, err := b.Block().HashTreeRoot()
if err != nil {
return err
}
oldBlockRoots[blkRoot] = true
if err := s.deleteBlockFromPendingQueue(slot, b, blkRoot); err != nil {
return err
}
}
}
}
return nil
}
func (s *Service) clearPendingSlots() {
s.pendingQueueLock.Lock()
defer s.pendingQueueLock.Unlock()
s.slotToPendingBlocks.Flush()
s.seenPendingBlocks = make(map[[32]byte]bool)
}
// Delete block from the list from the pending queue using the slot as key.
// Note: this helper is not thread safe.
func (s *Service) deleteBlockFromPendingQueue(slot primitives.Slot, b interfaces.ReadOnlySignedBeaconBlock, r [32]byte) error {
mutexasserts.AssertRWMutexLocked(&s.pendingQueueLock)
blks := s.pendingBlocksInCache(slot)
if len(blks) == 0 {
return nil
}
// Defensive check to ignore nil blocks
if err := blocks.BeaconBlockIsNil(b); err != nil {
return err
}
newBlks := make([]interfaces.ReadOnlySignedBeaconBlock, 0, len(blks))
for _, blk := range blks {
blkPb, err := blk.Proto()
if err != nil {
return err
}
bPb, err := b.Proto()
if err != nil {
return err
}
if equality.DeepEqual(blkPb, bPb) {
continue
}
newBlks = append(newBlks, blk)
}
if len(newBlks) == 0 {
s.slotToPendingBlocks.Delete(slotToCacheKey(slot))
delete(s.seenPendingBlocks, r)
return nil
}
// Decrease exp time in proportion to how many blocks are still in the cache for slot key.
d := pendingBlockExpTime / time.Duration(len(newBlks))
if err := s.slotToPendingBlocks.Replace(slotToCacheKey(slot), newBlks, d); err != nil {
return err
}
delete(s.seenPendingBlocks, r)
return nil
}
// Insert block to the list in the pending queue using the slot as key.
// Note: this helper is not thread safe.
func (s *Service) insertBlockToPendingQueue(_ primitives.Slot, b interfaces.ReadOnlySignedBeaconBlock, r [32]byte) error {
mutexasserts.AssertRWMutexLocked(&s.pendingQueueLock)
if s.seenPendingBlocks[r] {
return nil
}
if err := s.addPendingBlockToCache(b); err != nil {
return err
}
s.seenPendingBlocks[r] = true
return nil
}
// This returns signed beacon blocks given input key from slotToPendingBlocks.
func (s *Service) pendingBlocksInCache(slot primitives.Slot) []interfaces.ReadOnlySignedBeaconBlock {
k := slotToCacheKey(slot)
value, ok := s.slotToPendingBlocks.Get(k)
if !ok {
return []interfaces.ReadOnlySignedBeaconBlock{}
}
blks, ok := value.([]interfaces.ReadOnlySignedBeaconBlock)
if !ok {
return []interfaces.ReadOnlySignedBeaconBlock{}
}
return blks
}
// This adds input signed beacon block to slotToPendingBlocks cache.
func (s *Service) addPendingBlockToCache(b interfaces.ReadOnlySignedBeaconBlock) error {
if err := blocks.BeaconBlockIsNil(b); err != nil {
return err
}
blks := s.pendingBlocksInCache(b.Block().Slot())
if len(blks) >= maxBlocksPerSlot {
return nil
}
blks = append(blks, b)
k := slotToCacheKey(b.Block().Slot())
s.slotToPendingBlocks.Set(k, blks, pendingBlockExpTime)
return nil
}
// This converts input string to slot.
func cacheKeyToSlot(s string) primitives.Slot {
b := []byte(s)
return bytesutil.BytesToSlotBigEndian(b)
}
// This converts input slot to a key to be used for slotToPendingBlocks cache.
func slotToCacheKey(s primitives.Slot) string {
b := bytesutil.SlotToBytesBigEndian(s)
return string(b)
}
func dedupRoots(roots [][32]byte) [][32]byte {
newRoots := make([][32]byte, 0, len(roots))
rootMap := make(map[[32]byte]bool, len(roots))
for i, r := range roots {
if rootMap[r] {
continue
}
rootMap[r] = true
newRoots = append(newRoots, roots[i])
}
return newRoots
}