The GitHub Actions job "Flink Connector Kafka CI" on 
flink-connector-kafka.git/fix/FLINK-39712-no-more-split has failed.
Run started by GitHub user Savonitar (triggered by Savonitar).

Head commit for run:
c14889df377cd92631eccddfc8f3da9958fb6ed7 / Aleksandr Savonin 
<[email protected]>
[FLINK-39712][Connectors/Kafka] Avoid duplicate bounded-source completion 
notifications

A bounded DynamicKafkaSource can signal readers again after they have
finished when another reader registers or returns splits. An undeliverable
NoMoreSplitsEvent then fails the job. Record which readers have received
completion and re-arm the signal when a reader registers again.

Deduplication must be paired with assignment deferral: a cluster callback
can report completion before the remaining clusters assign their splits.
Defer completion across returned-split assignment, reader registration and
reported-split recovery, then recheck completion after each operation.
Keep the guard active through the recovery-gate drain because draining
clears its pending state before active and retained split assignment ends.
Centralize that synchronous guard in a private Runnable helper that clears
the flag in finally while preserving each completion callback's ordering.

Wire the real completion callback into the enumerator test context and add
five regressions covering startup, late registration, returned splits,
reported splits, and active plus retained reassignment. Assert both signal
counts and the exact split set present at completion.

The reader replay and recovery-gate prerequisites from #291 and #289 are
already on main. Metadata-change handling remains a follow-up in #309.

Generated-by: OpenAI Codex (GPT-6)

Report URL: 
https://github.com/apache/flink-connector-kafka/actions/runs/35531072502

With regards,
GitHub Actions via GitBox

Reply via email to