SEZ9 commented on PR #11932: URL: https://github.com/apache/seatunnel/pull/11932#issuecomment-5651014685
Quick status after re-checking the current head `b6c96b97356d`. **F7 (`Collector.restoreSchema` no-op on Flink/Spark)** — the Javadoc on the `restoreSchema` default method already states that only the Zeta engine collector restores this state today and that Flink/Spark keep the no-op until their translation-layer collectors support it. That is exactly the note I wanted so the connector-level restore in `IncrementalSourceReader` isn't misread as engine-agnostic, and it matches the text pointed to earlier in this thread (5578502768). I'm fine treating this as a known Zeta-only limitation tracked in a follow-up issue rather than a blocker for this `[Fix][Zeta]` PR — please link the follow-up issue here once it's filed and I'll mark F7 resolved. The remaining items are unchanged. Concrete asks: 1. **F1 / F2 (`IncrementalSourceReader`)** — the restore guard should not roll back to an empty schema when `checkpointTables` is non-null but empty (check for an empty list or fall back to the `checkpointDataType` condition), and the deserializer-side `restoreCheckpointProducedType` and the collector-side `restoreSchema` should be gated on the same condition so they can't diverge. 2. **F3 / F6 / F8 (`JdbcSink`)** — in `restoreWriter`, make the schema selection deterministic (or fail fast when merged states carry differing schemas), preserve the primary-key path when the restored schema lacks a `PrimaryKey` rather than silently going insert-only, and reject or filter null-Xid states before they reach `JdbcExactlyOnceSinkWriter` after an `exactly_once` config flip. 3. **F4 / F5 (`JdbcSinkState` / `JdbcSinkWriter`)** — add a docs/changelog note for the JDBC sink checkpoint state format change (schema persisted in `JdbcSinkState`, non-exactly-once writer now emits state), and confirm whether every `TableSchema`/column payload that can end up in that state is `Serializable`, or guard against it, so checkpoints that previously carried no writer state don't start failing. Once those land in a new commit, ping me and I'll do the full re-review. <!-- streview-comment:1020 --> -- 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]
