Setsushin commented on issue #17637: URL: https://github.com/apache/iceberg/issues/17637#issuecomment-6461799597
We hit this in production on 1.11.0 (12-task sink connector). Same fingerprint as the report: restarting one failed task (`POST /connectors/<name>/restart?includeTasks=true&onlyFailed=true`) triggered a rebalance. All 11 running tasks logged `found no partitions assigned across all members, cannot determine leader` within ~1 s, none logged `lost leader partition`, and 0.6 s later the restarted task logged `received leader partition`. The previous coordinator kept running for ~34 h, until we restarted all tasks of the connector. We also see this after node preemptions and node-pool upgrades, at times on several connectors at once. `CommitterImpl` is unchanged in 1.12.0, so the bug is still present there. +1 for the proposed shape (record the elected partition at `open()`, compare locally in `close()`). Our workaround meanwhile: restart all tasks of the connector instead of only the failed ones, because only the explicit-stop path (`close(List.of())`) reliably stops the coordinator. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
