ulysses-you commented on PR #58339: URL: https://github.com/apache/spark/pull/58339#issuecomment-5504387185
Thanks for the full re-derivation, @cloud-fan -- all ten findings are addressed in 7464c25; per-finding replies are on the threads. Two notes for the next pass: - **Interaction with SPARK-58968** (yours, merged while this round was open): it makes subset clustering correct for unmarked partitionings by inserting a projecting `GroupPartitionsExec`; this PR's `keysSatisfy` gate keeps unknown-keyed members out of that path -- every admission route (`satisfies`, `keysMaySatisfy`, `mayGroupToSatisfy`) funnels through `keysSatisfy`, whose marked branch demands full coverage. Rebased on top of it; both suites' tests are green together (marked subset => shuffle, unmarked subset => your projecting node, marked full-key => no shuffle, pinned by the positive control). - **Your review prompted a shuffle-count audit of our own tests**, which found two defects: an ORDER BY test that had silently degenerated to plain hash exchanges (a lost INSERT, caught by its new count assertion), and a master test line accidentally rewritten during the earlier rebase (restored). Both fixed. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
