xiangfu0 commented on PR #19711: URL: https://github.com/apache/pinot/pull/19711#issuecomment-6358961968
@gortiz thanks for flagging this. I checked #19954 against the rebased stack and found an additional worker-assignment gap: the ordered-leaf rewrite removes the final Sort, but its bounded merge receiver was not retaining the singleton requirement. At stageParallelism=2 on two servers, the regression test reproduced four merge workers instead of one in all three eligible cases, losing the random-server hotspot rotation. Fixed in 63eea61d11: bounded merge receivers request a singleton worker; unbounded receivers are unchanged. The inherited core already preserves partitioned workers for complete-input window sorts, including their compatibility MAX_VALUE encoding. I also extended the existing mixed window/final-ORDER-BY test to verify four window workers and one final-sort worker with both merge options enabled and disabled on a capable environment. All 279 QueryCompilationTest/DispatchablePlanVisitorTest cases pass on JDK 25, along with scoped formatting/license/checkstyle and warning-enabled compilation (zero added-line warnings). This follow-up changes only two files; no new test classes or benchmarks. I kept the broader future solution separate from this compatibility fix. The previous head's WindowFunnel CI assertion (expected 40 rows, got 36) also fails identically on its upstream base: [PR run](https://github.com/apache/pinot/actions/runs/39760676343/job/119763764275), [base run](https://github.com/apache/pinot/actions/runs/39742128873/job/119701663950). New-head hosted CI is a separate pending check, not covered by the local result above. -- 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]
