(fluss) branch main updated: [kv] Fix flaky KvTabletSnapshotTargetTest.testAddToSnapshotToStoreFail (#3264)

[yu...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

swuferhong pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/fluss.git


The following commit(s) were added to refs/heads/main by this push:
     new e3a48fd17 [kv] Fix flaky KvTabletSnapshotTargetTest.testAddToSnapshotToStoreFail (#3264)
e3a48fd17 is described below

commit e3a48fd173ef76607fe30f43bb93bdb744e6fd2e
Author: Kaixuan Duan <98...@users.noreply.github.com>
AuthorDate: Fri May 8 13:40:19 2026 +0800

    [kv] Fix flaky KvTabletSnapshotTargetTest.testAddToSnapshotToStoreFail (#3264)
---
 .../server/kv/snapshot/KvTabletSnapshotTarget.java | 28 +++++++++++++++-------
 1 file changed, 19 insertions(+), 9 deletions(-)

diff --git a/fluss-server/src/main/java/org/apache/fluss/server/kv/snapshot/KvTabletSnapshotTarget.java b/fluss-server/src/main/java/org/apache/fluss/server/kv/snapshot/KvTabletSnapshotTarget.java
index e3e8ede43..fe4591dc7 100644
--- a/fluss-server/src/main/java/org/apache/fluss/server/kv/snapshot/KvTabletSnapshotTarget.java
+++ b/fluss-server/src/main/java/org/apache/fluss/server/kv/snapshot/KvTabletSnapshotTarget.java
@@ -270,18 +270,28 @@ public class KvTabletSnapshotTarget implements PeriodicSnapshotManager.SnapshotT
     }
 
     /**
-     * Update local state after successful snapshot completion. This includes notifying RocksDB
-     * about completion, updating latest snapshot offset/size, and notifying LogTablet about the
-     * minimum offset to retain.
+     * Update local state after successful snapshot completion. This includes updating latest
+     * snapshot offset/size, notifying LogTablet about the minimum offset to retain, and finally
+     * notifying RocksDB about completion.
+     *
+     * <p>Order matters: outward-visible state (offset/size + {@code updateMinRetainOffset}
+     * callback) MUST be updated before {@link RocksIncrementalSnapshot#notifySnapshotComplete},
+     * otherwise observers waiting on {@code lastCompletedSnapshotId} may see the completion signal
+     * before the outward state catches up — see FLUSS-2624. The {@code finally} block also
+     * guarantees the internal SST bookkeeping cleanup runs even if any outward update throws.
      */
     private void updateStateOnCommitSuccess(long snapshotId, SnapshotResult snapshotResult) {
         long flushedLogOffset = snapshotResult.getTabletState().getFlushedLogOffset();
-        // notify the snapshot complete
-        rocksIncrementalSnapshot.notifySnapshotComplete(snapshotId);
-        logOffsetOfLatestSnapshot = flushedLogOffset;
-        snapshotSize = snapshotResult.getSnapshotSize();
-        // update LogTablet to notify the lowest offset that should be retained
-        updateMinRetainOffset.accept(flushedLogOffset);
+        try {
+            logOffsetOfLatestSnapshot = flushedLogOffset;
+            snapshotSize = snapshotResult.getSnapshotSize();
+            // update LogTablet to notify the lowest offset that should be retained
+            updateMinRetainOffset.accept(flushedLogOffset);
+        } finally {
+            // notify the snapshot complete (must run last to preserve the ordering contract above,
+            // and run in finally to guarantee internal SST bookkeeping is always cleaned up).
+            rocksIncrementalSnapshot.notifySnapshotComplete(snapshotId);
+        }
     }
 
     /**