SEZ9 commented on PR #13092: URL: https://github.com/apache/seatunnel/pull/13092#issuecomment-6889200084
Thanks for the update. The automated review notes that the fork Build (run 43368787049) and the Apache Build now pass on the unchanged head 0b00191d31165c1a48ba5622fb7837a436867e74, so CI is no longer the blocker here. What remains is what @nzw921rx flagged: the documentation and the nullability helper. Concretely, before this can go in: 1. **Nullability gate (F1 / F7)** – `isNullableSplitKey` currently decides from catalog metadata, and that metadata describes the base table, not an arbitrary `query` projection (outer-joined NOT NULL columns come back as NULL). Please make the automatic path only trust metadata when the split key comes from a real table, and otherwise treat the key as nullable (probe + NULL split). Also add Javadoc to `isNullableSplitKey` stating that it is an extension point and why the `orElse(true)` fallback is the chosen policy. The "automatic choice vs. explicit partition column" contract should be spelled out in that Javadoc and in the docs, so the two code paths are clearly distinguished. 2. **Docs (F2 / F6)** – In the "NULL values in the split column" section of `docs/en/connectors/source/Jdbc.md`, please describe how the extra NULL split interacts with explicit `partition_lower_bound` / `partition_upper_bound` (the troubleshooting section still says those define the read range). In `docs/en/introduction/concepts/incompatible-changes.md`, please correct the "same splits as before" statement: hash splits on a nullable column now carry an extra `AND <col> IS NOT NULL` predicate even when no NULLs exist. 3. **Still open from the earlier round (F3 / F4 / F5 / F8)** – a short answer on each is enough: (a) whether the NULL probe should fail planning on any `SQLException` or degrade gracefully, (b) whether the probe can ride on the existing MIN/MAX scan for range splits, (c) a note in the incompatible-changes doc that a checkpoint containing the NULL split becomes a full-table split if restored with the previous jar, and (d) a unit test pinning old-split → new-class deserialization instead of the manual verification. Once 1 and 2 are pushed and 3 is answered or addressed, I'll re-run the review on the new head. <!-- streview-comment:2371 --> -- 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]
