From ea7d6e21403946bbf05ae8e327eed335bee726ac Mon Sep 17 00:00:00 2001
From: zenground0 <ZenGround0@users.noreply.github.com>
Date: Thu, 23 Feb 2023 18:34:26 +0530
Subject: [PATCH] Two fixes, better logging and comments

---
 blockstore/splitstore/markset_badger.go     |  2 +-
 blockstore/splitstore/splitstore_compact.go | 17 ++++++++++-------
 blockstore/splitstore/splitstore_prune.go   |  4 ++--
 3 files changed, 13 insertions(+), 10 deletions(-)

diff --git a/blockstore/splitstore/markset_badger.go b/blockstore/splitstore/markset_badger.go
index 8808a621709..2dac673cd76 100644
--- a/blockstore/splitstore/markset_badger.go
+++ b/blockstore/splitstore/markset_badger.go
@@ -351,7 +351,7 @@ func (s *BadgerMarkSet) write(seqno int) (err error) {
 	persist := s.persist
 	s.mx.RUnlock()
 
-	if persist && !system.BadgerFsyncDisable {
+	if persist && !system.BadgerFsyncDisable { // WARNING: disabling sync makes recovery from crash during critical section unsound
 		return s.db.Sync()
 	}
 
diff --git a/blockstore/splitstore/splitstore_compact.go b/blockstore/splitstore/splitstore_compact.go
index 0422736f03c..4e82f55bc84 100644
--- a/blockstore/splitstore/splitstore_compact.go
+++ b/blockstore/splitstore/splitstore_compact.go
@@ -115,8 +115,6 @@ func (s *SplitStore) HeadChange(_, apply []*types.TipSet) error {
 		return nil
 	}
 
-	// Prioritize hot store compaction over cold store prune
-
 	if epoch-s.baseEpoch > CompactionThreshold {
 		// it's time to compact -- prepare the transaction and go!
 		s.beginTxnProtect()
@@ -176,6 +174,8 @@ func (s *SplitStore) protectTipSets(apply []*types.TipSet) {
 		timestamp := time.Unix(int64(curTs.MinTimestamp()), 0)
 		doSync := time.Since(timestamp) < SyncWaitTime
 		go func() {
+			// we are holding the txnLk while marking
+			// so critical section cannot delete
 			if doSync {
 				defer func() {
 					s.txnSyncMx.Lock()
@@ -689,7 +689,7 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 
 	log.Infow("cold collection done", "took", time.Since(startCollect))
 
-	log.Infow("compaction stats", "hot", hotCnt, "cold", coldCnt)
+	log.Infow("compaction stats", "hot", hotCnt, "cold", coldCnt, "purge", purgeCnt)
 	stats.Record(s.ctx, metrics.SplitstoreCompactionHot.M(int64(hotCnt)))
 	stats.Record(s.ctx, metrics.SplitstoreCompactionCold.M(int64(coldCnt)))
 
@@ -788,6 +788,9 @@ func (s *SplitStore) doCompact(curTs *types.TipSet) error {
 	if err := os.Remove(s.coldSetPath()); err != nil {
 		log.Warnf("error removing coldset: %s", err)
 	}
+	if err := os.Remove(s.discardSetPath()); err != nil {
+		log.Warnf("error removing discardset: %s", err)
+	}
 
 	// we are done; do some housekeeping
 	s.endTxnProtect()
@@ -939,8 +942,8 @@ func (s *SplitStore) walkChain(ts *types.TipSet, inclState, inclMsgs abi.ChainEp
 		if err != nil {
 			return xerrors.Errorf("error computing cid reference to parent tipset")
 		}
-		if err := s.walkObjectIncomplete(pRef, visitor, fHot, stopWalk); err != nil {
-			return xerrors.Errorf("error walking parent tipset cid reference")
+		if err := fHot(pRef); err != nil {
+			return xerrors.Errorf("error marking parent tipset cid reference")
 		}
 
 		// message are retained if within the inclMsgs boundary
@@ -998,8 +1001,8 @@ func (s *SplitStore) walkChain(ts *types.TipSet, inclState, inclMsgs abi.ChainEp
 	if err != nil {
 		return xerrors.Errorf("error computing cid reference to parent tipset")
 	}
-	if err := s.walkObjectIncomplete(hRef, visitor, fHot, stopWalk); err != nil {
-		return xerrors.Errorf("error walking parent tipset cid reference")
+	if err := fHot(hRef); err != nil {
+		return xerrors.Errorf("error marking parent tipset cid reference")
 	}
 
 	for len(toWalk) > 0 {
diff --git a/blockstore/splitstore/splitstore_prune.go b/blockstore/splitstore/splitstore_prune.go
index 6a26c00d271..4a38045e8c6 100644
--- a/blockstore/splitstore/splitstore_prune.go
+++ b/blockstore/splitstore/splitstore_prune.go
@@ -329,9 +329,9 @@ func (s *SplitStore) doPrune(curTs *types.TipSet, retainStateP func(int64) bool,
 	}
 
 	s.pruneIndex++
-	err = s.ds.Put(s.ctx, pruneIndexKey, int64ToBytes(s.compactionIndex))
+	err = s.ds.Put(s.ctx, pruneIndexKey, int64ToBytes(s.pruneIndex))
 	if err != nil {
-		return xerrors.Errorf("error saving compaction index: %w", err)
+		return xerrors.Errorf("error saving prune index: %w", err)
 	}
 
 	return nil