optimize(op-node): make block produce stable when L1 latency unstable (…

…ethereum-optimism#87)

* optimize(op-node): make block produce stable when L1 latency unstable

* fix find origin logic

* fix origin_selector_test

* fix ci

* fix ci

---------

Co-authored-by: Owen <[email protected]>

op-node/rollup/driver/origin_selector.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"time"
 
 	"github.com/ethereum/go-ethereum"
 	"github.com/ethereum/go-ethereum/log"
@@ -16,6 +17,7 @@ import (
 type L1Blocks interface {
 	derive.L1BlockRefByHashFetcher
 	derive.L1BlockRefByNumberFetcher
+	derive.L1ReceiptsFetcher
 }
 
 type L1OriginSelector struct {
@@ -48,14 +50,19 @@ func (los *L1OriginSelector) FindL1Origin(ctx context.Context, l2Head eth.L2Bloc
 	// If we are past the sequencer depth, we may want to advance the origin, but need to still
 	// check the time of the next origin.
 	pastSeqDrift := l2Head.Time+los.cfg.BlockTime > currentOrigin.Time+los.cfg.MaxSequencerDrift
+	// Limit the time to fetch next origin block by default
+	refCtx, refCancel := context.WithTimeout(ctx, 100*time.Millisecond)
+	defer refCancel()
 	if pastSeqDrift {
 		log.Warn("Next L2 block time is past the sequencer drift + current origin time")
+		// Must fetch next L1 block as long as it may take, cause we are pastSeqDrift
+		refCtx = ctx
 	}
 
 	// Attempt to find the next L1 origin block, where the next origin is the immediate child of
 	// the current origin block.
 	// The L1 source can be shimmed to hide new L1 blocks and enforce a sequencer confirmation distance.
-	nextOrigin, err := los.l1.L1BlockRefByNumber(ctx, currentOrigin.Number+1)
+	nextOrigin, err := los.l1.L1BlockRefByNumber(refCtx, currentOrigin.Number+1)
 	if err != nil {
 		if pastSeqDrift {
 			return eth.L1BlockRef{}, fmt.Errorf("cannot build next L2 block past current L1 origin %s by more than sequencer time drift, and failed to find next L1 origin: %w", currentOrigin, err)
@@ -68,12 +75,21 @@ func (los *L1OriginSelector) FindL1Origin(ctx context.Context, l2Head eth.L2Bloc
 		return currentOrigin, nil
 	}
 
+	receiptsCached := true
+	receiptsCtx, receiptsCancel := context.WithTimeout(ctx, 10*time.Millisecond)
+	defer receiptsCancel()
+	_, _, err = los.l1.FetchReceipts(receiptsCtx, nextOrigin.Hash)
+	if err != nil {
+		receiptsCached = false
+	}
+
 	// If the next L2 block time is greater than the next origin block's time, we can choose to
 	// start building on top of the next origin. Sequencer implementation has some leeway here and
 	// could decide to continue to build on top of the previous origin until the Sequencer runs out
 	// of slack. For simplicity, we implement our Sequencer to always start building on the latest
 	// L1 block when we can.
-	if l2Head.Time+los.cfg.BlockTime >= nextOrigin.Time {
+	// If not pastSeqDrift and next origin receipts not cached, fallback to current origin.
+	if l2Head.Time+los.cfg.BlockTime >= nextOrigin.Time && (pastSeqDrift || receiptsCached) {
 		return nextOrigin, nil
 	}
 

op-node/rollup/driver/origin_selector_test.go

@@ -46,6 +46,7 @@ func TestOriginSelectorAdvances(t *testing.T) {
 
 	l1.ExpectL1BlockRefByHash(a.Hash, a, nil)
 	l1.ExpectL1BlockRefByNumber(b.Number, b, nil)
+	l1.ExpectFetchReceipts(b.Hash, nil, nil, nil)
 
 	s := NewL1OriginSelector(log, cfg, l1)
 	next, err := s.FindL1Origin(context.Background(), l2Head)
@@ -87,6 +88,7 @@ func TestOriginSelectorRespectsOriginTiming(t *testing.T) {
 
 	l1.ExpectL1BlockRefByHash(a.Hash, a, nil)
 	l1.ExpectL1BlockRefByNumber(b.Number, b, nil)
+	l1.ExpectFetchReceipts(b.Hash, nil, nil, nil)
 
 	s := NewL1OriginSelector(log, cfg, l1)
 	next, err := s.FindL1Origin(context.Background(), l2Head)
@@ -208,6 +210,7 @@ func TestOriginSelectorSeqDriftRespectsNextOriginTime(t *testing.T) {
 
 	l1.ExpectL1BlockRefByHash(a.Hash, a, nil)
 	l1.ExpectL1BlockRefByNumber(b.Number, b, nil)
+	l1.ExpectFetchReceipts(b.Hash, nil, nil, nil)
 
 	s := NewL1OriginSelector(log, cfg, l1)
 	next, err := s.FindL1Origin(context.Background(), l2Head)
@@ -265,6 +268,7 @@ func TestOriginSelectorHandlesLateL1Blocks(t *testing.T) {
 	l1.ExpectL1BlockRefByHash(a.Hash, a, nil)
 	l1.ExpectL1BlockRefByHash(a.Hash, a, nil)
 	l1.ExpectL1BlockRefByNumber(b.Number, b, nil)
+	l1.ExpectFetchReceipts(b.Hash, nil, nil, nil)
 
 	l1Head := b
 	confDepthL1 := NewConfDepth(2, func() eth.L1BlockRef { return l1Head }, l1)

op-node/rollup/driver/state.go

@@ -333,7 +333,13 @@ func (s *Driver) eventLoop() {
 		case <-stepReqCh:
 			s.metrics.SetDerivationIdle(false)
 			s.log.Debug("Derivation process step", "onto_origin", s.derivation.Origin(), "attempts", stepAttempts)
-			err := s.derivation.Step(context.Background())
+			stepCtx := context.Background()
+			if s.driverConfig.SequencerEnabled && !s.driverConfig.SequencerStopped {
+				var cancelStep context.CancelFunc
+				stepCtx, cancelStep = context.WithTimeout(ctx, 3*time.Second)
+				defer cancelStep()
+			}
+			err := s.derivation.Step(stepCtx)
 			stepAttempts += 1 // count as attempt by default. We reset to 0 if we are making healthy progress.
 			if err == io.EOF {
 				s.log.Debug("Derivation process went idle", "progress", s.derivation.Origin(), "err", err)

op-node/sources/l1_client.go

@@ -27,8 +27,8 @@ func L1ClientDefaultConfig(config *rollup.Config, trustRPC bool, kind RPCProvide
 	// Cache 3/2 worth of sequencing window of receipts and txs
 	span := int(config.SeqWindowSize) * 3 / 2
 	fullSpan := span
-	if span > 1000 { // sanity cap. If a large sequencing window is configured, do not make the cache too large
-		span = 1000
+	if span > 2000 { // sanity cap. If a large sequencing window is configured, do not make the cache too large
+		span = 2000
 	}
 	return &L1ClientConfig{
 		EthClientConfig: EthClientConfig{
@@ -132,6 +132,7 @@ func (s *L1Client) L1BlockRefByHash(ctx context.Context, hash common.Hash) (eth.
 }
 
 func (s *L1Client) GoOrUpdatePreFetchReceipts(ctx context.Context, l1Start uint64) error {
+	s.preFetchReceiptsStartBlockChan <- l1Start
 	s.preFetchReceiptsOnce.Do(func() {
 		s.log.Info("pre-fetching receipts start", "startBlock", l1Start)
 		go func() {
@@ -162,7 +163,6 @@ func (s *L1Client) GoOrUpdatePreFetchReceipts(ctx context.Context, l1Start uint6
 			}
 		}()
 	})
-	s.preFetchReceiptsStartBlockChan <- l1Start
 	return nil
 }