SEZ9 commented on PR #11077: URL: https://github.com/apache/seatunnel/pull/11077#issuecomment-6870359301
Thanks for the update on `352e5a27`. Going back through the earlier points, a few confirmations would help: **F1 (destination-key collisions routing data across sinks)** – Still the main item for me. If the destination key is derived from a hash of the options, two genuinely different destinations can collapse into one writer. Please either make the key a stable, collision-free representation of the normalized destination, or compare the full identity on collision and fail fast instead of sharing the writer. **F2 (snapshot fan-out / state duplicated on restore)** – Could you confirm that shared-writer state is now persisted once under a single canonical identifier? If the table list is reordered or an alias is removed between runs, does the restore path still locate that state rather than silently dropping it? **F3 (shared writer built from an arbitrary alias)** – Is there validation that aliases sharing a destination key have identical physical row layouts? If so, what happens when they differ, i.e. does the job fail at `MultiTableSink` construction with a message naming the tables involved? **F5 (`proxyContexts` registering only the first alias, `containsValue` scan)** – Please confirm whether every alias now gets a context entry and whether the `containsValue` scan was replaced with a lookup keyed by destination. **F7 (IOException wrapped in `RuntimeException` inside `computeIfAbsent`)** – Please confirm whether writer creation/restoration still goes through `computeIfAbsent` and, if not, that `IOException` from `createWriter`/`restoreWriter` now propagates as declared. **F4 / F6 (docs for `getPhysicalDestinationIdentifier()` and the changed `restoreWriter` contract)** – Could you point me to where the new SPI method and the merged-state `restoreWriter` behavior are described? **F8 (`getDestinationKey` Javadoc)** – Minor; please add the parameter and return tags if not already done. Once F1, F5 and F7 are confirmed or addressed, I think this is close. <!-- streview-comment:2334 --> -- 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]
