Merge pull request #10641 from filecoin-project/phi/backport-splitstore-oos

Backport #10392 into v1.21.0

blockstore/badger/blockstore.go

@@ -444,7 +444,7 @@ func (b *Blockstore) deleteDB(path string) {
 	}
 }
 
-func (b *Blockstore) onlineGC(ctx context.Context, threshold float64) error {
+func (b *Blockstore) onlineGC(ctx context.Context, threshold float64, checkFreq time.Duration, check func() error) error {
 	b.lockDB()
 	defer b.unlockDB()
 
@@ -461,11 +461,15 @@ func (b *Blockstore) onlineGC(ctx context.Context, threshold float64) error {
 	if err != nil {
 		return err
 	}
-
+	checkTick := time.NewTimer(checkFreq)
+	defer checkTick.Stop()
 	for err == nil {
 		select {
 		case <-ctx.Done():
 			err = ctx.Err()
+		case <-checkTick.C:
+			err = check()
+			checkTick.Reset(checkFreq)
 		default:
 			err = b.db.RunValueLogGC(threshold)
 		}
@@ -502,7 +506,17 @@ func (b *Blockstore) CollectGarbage(ctx context.Context, opts ...blockstore.Bloc
 	if threshold == 0 {
 		threshold = defaultGCThreshold
 	}
-	return b.onlineGC(ctx, threshold)
+	checkFreq := options.CheckFreq
+	if checkFreq < 30*time.Second { // disallow checking more frequently than block time
+		checkFreq = 30 * time.Second
+	}
+	check := options.Check
+	if check == nil {
+		check = func() error {
+			return nil
+		}
+	}
+	return b.onlineGC(ctx, threshold, checkFreq, check)
 }
 
 // GCOnce runs garbage collection on the value log;

blockstore/blockstore.go

@@ -2,6 +2,7 @@ package blockstore
 
 import (
 	"context"
+	"time"
 
 	"github.com/ipfs/go-cid"
 	ds "github.com/ipfs/go-datastore"
@@ -57,6 +58,10 @@ type BlockstoreGCOptions struct {
 	FullGC bool
 	// fraction of garbage in badger vlog before its worth processing in online GC
 	Threshold float64
+	// how often to call the check function
+	CheckFreq time.Duration
+	// function to call periodically to pause or early terminate GC
+	Check func() error
 }
 
 func WithFullGC(fullgc bool) BlockstoreGCOption {
@@ -73,6 +78,20 @@ func WithThreshold(threshold float64) BlockstoreGCOption {
 	}
 }
 
+func WithCheckFreq(f time.Duration) BlockstoreGCOption {
+	return func(opts *BlockstoreGCOptions) error {
+		opts.CheckFreq = f
+		return nil
+	}
+}
+
+func WithCheck(check func() error) BlockstoreGCOption {
+	return func(opts *BlockstoreGCOptions) error {
+		opts.Check = check
+		return nil
+	}
+}
+
 // BlockstoreSize is a trait for on-disk blockstores that can report their size
 type BlockstoreSize interface {
 	Size() (int64, error)

blockstore/splitstore/splitstore.go

@@ -187,6 +187,11 @@ type SplitStore struct {
 	ctx    context.Context
 	cancel func()
 
+	outOfSync         int32 // for fast checking
+	chainSyncMx       sync.Mutex
+	chainSyncCond     sync.Cond
+	chainSyncFinished bool // protected by chainSyncMx
+
 	debug *debugLog
 
 	// transactional protection for concurrent read/writes during compaction
@@ -261,6 +266,7 @@ func Open(path string, ds dstore.Datastore, hot, cold bstore.Blockstore, cfg *Co
 
 	ss.txnViewsCond.L = &ss.txnViewsMx
 	ss.txnSyncCond.L = &ss.txnSyncMx
+	ss.chainSyncCond.L = &ss.chainSyncMx
 	ss.ctx, ss.cancel = context.WithCancel(context.Background())
 
 	ss.reifyCond.L = &ss.reifyMx
@@ -822,6 +828,11 @@ func (s *SplitStore) Close() error {
 		s.txnSyncCond.Broadcast()
 		s.txnSyncMx.Unlock()
 
+		s.chainSyncMx.Lock()
+		s.chainSyncFinished = true
+		s.chainSyncCond.Broadcast()
+		s.chainSyncMx.Unlock()
+
 		log.Warn("close with ongoing compaction in progress; waiting for it to finish...")
 		for atomic.LoadInt32(&s.compacting) == 1 {
 			time.Sleep(time.Second)

blockstore/splitstore/splitstore_compact.go

@@ -91,7 +91,35 @@ func (s *SplitStore) HeadChange(_, apply []*types.TipSet) error {
 	//      Regardless, we put a mutex in HeadChange just to be safe
 
 	if !atomic.CompareAndSwapInt32(&s.compacting, 0, 1) {
-		// we are currently compacting -- protect the new tipset(s)
+		// we are currently compacting
+		// 1. Signal sync condition to yield compaction when out of sync and resume when in sync
+		timestamp := time.Unix(int64(curTs.MinTimestamp()), 0)
+		if CheckSyncGap && time.Since(timestamp) > SyncGapTime {
+			/* Chain out of sync */
+			if atomic.CompareAndSwapInt32(&s.outOfSync, 0, 1) {
+				// transition from in sync to out of sync
+				s.chainSyncMx.Lock()
+				s.chainSyncFinished = false
+				s.chainSyncMx.Unlock()
+			}
+			// already out of sync, no signaling necessary
+
+		}
+		// TODO: ok to use hysteresis with no transitions between 30s and 1m?
+		if time.Since(timestamp) < SyncWaitTime {
+			/* Chain in sync */
+			if atomic.CompareAndSwapInt32(&s.outOfSync, 0, 0) {
+				// already in sync, no signaling necessary
+			} else {
+				// transition from out of sync to in sync
+				s.chainSyncMx.Lock()
+				s.chainSyncFinished = true
+				s.chainSyncCond.Broadcast()
+				s.chainSyncMx.Unlock()
+			}
+
+		}
+		// 2. protect the new tipset(s)
 		s.protectTipSets(apply)
 		return nil
 	}
@@ -427,7 +455,7 @@ func (s *SplitStore) protectTxnRefs(markSet MarkSet) error {
 // transactionally protect a reference by walking the object and marking.
 // concurrent markings are short circuited by checking the markset.
 func (s *SplitStore) doTxnProtect(root cid.Cid, markSet MarkSet) (int64, error) {
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return 0, err
 	}
 
@@ -545,7 +573,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 	}
 	defer coldSet.Close() //nolint:errcheck
 
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return err
 	}
 
@@ -617,7 +645,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 
 	log.Infow("marking done", "took", time.Since(startMark), "marked", *count)
 
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return err
 	}
 
@@ -627,7 +655,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 		return xerrors.Errorf("error protecting transactional refs: %w", err)
 	}
 
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return err
 	}
 
@@ -704,7 +732,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 	stats.Record(s.ctx, metrics.SplitstoreCompactionHot.M(hotCnt))
 	stats.Record(s.ctx, metrics.SplitstoreCompactionCold.M(coldCnt))
 
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return err
 	}
 
@@ -713,7 +741,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 	// possibly delete objects we didn't have when we were collecting cold objects)
 	s.waitForMissingRefs(markSet)
 
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return err
 	}
 
@@ -733,7 +761,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 		}
 		log.Infow("moving done", "took", time.Since(startMove))
 
-		if err := s.checkClosing(); err != nil {
+		if err := s.checkYield(); err != nil {
 			return err
 		}
 
@@ -764,7 +792,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 	}
 
 	// wait for the head to catch up so that the current tipset is marked
-	s.waitForSync()
+	s.waitForTxnSync()
 
 	if err := s.checkClosing(); err != nil {
 		return err
@@ -865,7 +893,7 @@ func (s *SplitStore) beginCriticalSection(markSet MarkSet) error {
 	return nil
 }
 
-func (s *SplitStore) waitForSync() {
+func (s *SplitStore) waitForTxnSync() {
 	log.Info("waiting for sync")
 	if !CheckSyncGap {
 		log.Warnf("If you see this outside of test it is a serious splitstore issue")
@@ -884,6 +912,25 @@ func (s *SplitStore) waitForSync() {
 	}
 }
 
+// Block compaction operations if chain sync has fallen behind
+func (s *SplitStore) waitForSync() {
+	if atomic.LoadInt32(&s.outOfSync) == 0 {
+		return
+	}
+	s.chainSyncMx.Lock()
+	defer s.chainSyncMx.Unlock()
+
+	for !s.chainSyncFinished {
+		s.chainSyncCond.Wait()
+	}
+}
+
+// Combined sync and closing check
+func (s *SplitStore) checkYield() error {
+	s.waitForSync()
+	return s.checkClosing()
+}
+
 func (s *SplitStore) endTxnProtect() {
 	s.txnLk.Lock()
 	defer s.txnLk.Unlock()
@@ -1037,7 +1084,7 @@ func (s *SplitStore) walkChain(ts *types.TipSet, inclState, inclMsgs abi.ChainEp
 
 	for len(toWalk) > 0 {
 		// walking can take a while, so check this with every opportunity
-		if err := s.checkClosing(); err != nil {
+		if err := s.checkYield(); err != nil {
 			return err
 		}
 
@@ -1106,7 +1153,7 @@ func (s *SplitStore) walkObject(c cid.Cid, visitor ObjectVisitor, f func(cid.Cid
 	}
 
 	// check this before recursing
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return 0, err
 	}
 
@@ -1175,7 +1222,7 @@ func (s *SplitStore) walkObjectIncomplete(c cid.Cid, visitor ObjectVisitor, f, m
 	}
 
 	// check this before recursing
-	if err := s.checkClosing(); err != nil {
+	if err := s.checkYield(); err != nil {
 		return sz, err
 	}
 
@@ -1262,7 +1309,7 @@ func (s *SplitStore) moveColdBlocks(coldr *ColdSetReader) error {
 	batch := make([]blocks.Block, 0, batchSize)
 
 	err := coldr.ForEach(func(c cid.Cid) error {
-		if err := s.checkClosing(); err != nil {
+		if err := s.checkYield(); err != nil {
 			return err
 		}
 		blk, err := s.hot.Get(s.ctx, c)

blockstore/splitstore/splitstore_gc.go

@@ -65,10 +65,15 @@ func (s *SplitStore) gcHotAfterCompaction() {
 }
 
 func (s *SplitStore) gcBlockstore(b bstore.Blockstore, opts []bstore.BlockstoreGCOption) error {
+	if err := s.checkYield(); err != nil {
+		return err
+	}
 	if gc, ok := b.(bstore.BlockstoreGC); ok {
 		log.Info("garbage collecting blockstore")
 		startGC := time.Now()
 
+		opts = append(opts, bstore.WithCheckFreq(90*time.Second))
+		opts = append(opts, bstore.WithCheck(s.checkYield))
 		if err := gc.CollectGarbage(s.ctx, opts...); err != nil {
 			return err
 		}