The GitHub Actions job "Nightly (beta)" on flink.git/master has failed.
Run started by GitHub user github-actions[bot] (triggered by 
github-actions[bot]).

Head commit for run:
2f3c205e9266cb30240eb7f4fdab15cad629a70f / Roman Khachatryan 
<[email protected]>
[FLINK-40081][test] Add CdrRecoveryRaceITCase for the non-source recovery race

Adds a bounded BATCH job (blocking exchange -> non-source operator) that
exercises the checkpointing-during-recovery restore path for an operator
whose input is fully available, i.e. the scenario that tripped "Mailbox
loop interrupted before recovery was finished" on TPC-DS and is fixed by
the defer-task-finish commit of this series.

INCOMPLETE -- does not reliably guard the regression. The race window on
a fresh deployment with no recovered channel state is tiny, so this test
does NOT deterministically reproduce the bug on its own. During
development it reproduced only with a temporary delay injected into the
recovery chain to widen the window; that delay is not product code and
is not included here. As committed it is a fast BATCH + CDR smoke test
of the no-recovered-state recovery path, not a guaranteed reproducer.

Follow-up (see TODO in the test): make it deterministic via a test-only
hook to widen the recovery window, or an unaligned-checkpoint restore
that carries in-flight channel state, before treating it as a real
regression test.

Report URL: https://github.com/apache/flink/actions/runs/31764880735

With regards,
GitHub Actions via GitBox

Reply via email to