DanielLeens commented on issue #12348: URL: https://github.com/apache/seatunnel/issues/12348#issuecomment-6217826347
Classification: A / Connector-V2 MySQL-CDC savepoint recovery schema bootstrap. Thank you for the careful 3.0.0 reproduction. The supplied trace establishes the failure before any sink assertion is needed: the enumerator recognizes table B as snapshot work, while the restored reader starts the A-only incremental split and then fails when B's TABLE_MAP event arrives without schema history. I rechecked the current dev source and related work. There is no merged fix or open PR covering this exact configuration-change plus restore contract. #11206 concerns a different live-discovery path and is still open with a failed Build; the earlier #12567 PR is closed. The exactly-once=false line in this reproduction is important: it means this report proves the schema-bootstrap failure, but does not yet prove checkpoint-boundary delivery semantics. Please keep a fix focused on the state migration. It must initialize B's schema/history and snapshot-consistent watermark before the restored incremental reader can consume B events; suppressing the TABLE_MAP event or merely adding B to an incremental split would risk loss. The acceptance test should cover both savepoint and checkpoint restore: restore A from its saved offset, add B, verify B receives exactly one initial snapshot at a defined watermark, then verify B changes are consumed while A neither re-snapshots nor loses or replays its restored stream. Include an assertion for the configuration where connector exactly-once is actually enabled, and keep the currently demonstrated non-exactly-once failure as a separate regression case if its contract differs. -- 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]
