Open tbg opened 1 year ago
cc @cockroachdb/replication
Reading the master
code, I was wrong - there is no bug, I introduced (and caught) it on https://github.com/cockroachdb/cockroach/pull/97779. But the code is pretty brittle and the test coverage is missing so leaving this issue open to track that.
Hi @erikgrinaker, please add branch-* labels to identify which branch(es) this release-blocker affects.
:owl: Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf.
Marking a release blocker to make sure we add test coverage and catch any lingering bugs here before the release.
We won't have time to get to this, dropping GA-blocker.
When, during proposal application, we attempt a reproposal but this fails, we signal the proposal with a (final) error. However, we fail to clear the local side effects. I think this can cause intents to be resolved that aren't committed yet. TestFailureToProcessCommandClearsLocalResult` exercises this general path, except not with the failing reproposal - it lets the reproposal succeed and simply verifies that the original command's local result is cleared. But if the reproposal fails, a different path is taken and the local result is not cleared, but it ought to be.
We should update our test coverage and fix this bug.
Found in #97779
Jira issue: CRDB-28955
Epic CRDB-39898