SEZ9 commented on PR #11932: URL: https://github.com/apache/seatunnel/pull/11932#issuecomment-5643221978
Thanks for the update on `b6c96b97356d`. Treating Flink/Spark `Collector.restoreSchema` as a known Zeta-only limitation tracked in a follow-up issue is fine with me for this `[Fix][Zeta]`-scoped PR, as long as the default method in `seatunnel-api/src/main/java/org/apache/seatunnel/api/source/Collector.java` gets a short Javadoc note saying it is currently only honoured by the Zeta engine, so the connector-level restore in `IncrementalSourceReader` isn't read as working everywhere. Please link the follow-up issue here once it exists. What I still need before another full pass: 1. **PR11932-F1 / F2 (`IncrementalSourceReader`)** – confirm the schema-restore guard still handles a non-null-but-empty `checkpointTables` without rolling the deserializer/collector back to an empty schema, and that `restoreCheckpointProducedType` and the `getSchemaChangeResolver() != null`-gated collector restore can't diverge. A short note on how you verified this (or a unit test) is enough. 2. **PR11932-F3 / F6 / F8 (`JdbcSink`)** – tell me how `restoreWriter` behaves when merged states carry differing `TableSchema`s (first-non-null pick), when a restored schema has no `PrimaryKey` (silent insert-only downgrade), and when null-Xid states from `JdbcSinkWriter` reach `JdbcExactlyOnceSinkWriter` after an `exactly_once` flip. If any of these are intentionally out of scope, say so explicitly and I'll consider a follow-up. 3. **PR11932-F4 / F5 (`JdbcSinkState` / `JdbcSinkWriter`)** – the docs/upgrade verification is still open: please add a docs or changelog note covering the new persisted schema in `JdbcSinkState` and that the non-exactly-once writer now emits state, and confirm whether a non-Serializable column payload can fail checkpoints that previously carried no writer state. Once those are addressed or answered, ping here and I'll re-review. <!-- streview-comment:995 --> -- 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]
