SEZ9 commented on PR #13092: URL: https://github.com/apache/seatunnel/pull/13092#issuecomment-6841794694
Thanks for the detailed follow-up on `96594e52d5` — the four points line up with the first four findings, and the approach on each sounds right to me: 1. **Query-based nullability**: treating any `query` source as nullable (always run the NULL check plus the hash `IS NOT NULL`) while keeping schema trust for table paths is the split I was hoping for. The test with a NOT NULL schema but a NULL-returning query, covering both range and hash with every id read once, is exactly the regression I wanted pinned. 2. **Explicit bounds**: agreed that with both numeric bounds the fixed splitter only reads closed ranges inside them, so NULL is legitimately outside the range like any other out-of-range value, and `readsOnlyConfiguredRange` overridden in `FixedChunkSplitter` expresses that cleanly. Keeping the NULL split for string keys makes sense given their hash splits ignore the bounds. Thanks for updating Jdbc.md (EN and ZH) and the troubleshooting sentence. 3. **Probe failure → warn and add the NULL split anyway**: that's the right failure mode (correctness over a planning abort), and the failing-`createStatement` connection test covers it. 4. **Separate probe vs. riding on MIN/MAX**: your reasoning is convincing — turning an index-answered MIN/MAX into a full scan via `SUM(CASE WHEN c IS NULL ...)` would be a worse trade than a second cheap probe. I'm fine keeping it separate; a short code comment explaining why would help the next reader, but it's not blocking. Your comment appears to be cut off at "the column picked automatically is a uniqu…", so I may have missed the tail. Could you also confirm where things stand on the remaining review items? - **Rollback compatibility** (a checkpoint containing the new NULL split becoming a full-table split when restored with the previous connector jar): addressed, or do you intend to document it as a known upgrade caveat? - **Incompatible-changes note**: has the "same splits as before" wording been adjusted to mention the extra `AND <col> IS NOT NULL` predicate on hash splits of nullable columns? - **`isNullableSplitKey` Javadoc**: with the new query-aware behaviour this is a fairly nuanced extension point; please add Javadoc covering the `query` rule and the `orElse(true)` fallback policy. - **Old-split → new-class deserialization test**: is there now a unit test pinning this, or is it still manual-only? Once those are confirmed (or the truncated part of your comment is re-posted), I'll do a final pass. <!-- streview-comment:2306 --> -- 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]
