DanielLeens commented on issue #12058: URL: https://github.com/apache/seatunnel/issues/12058#issuecomment-5679935939
I reviewed the linked wait-site report and its raw JMH, profiler and zero-threshold JFR artifacts. The evidence now supports a narrow conclusion: the `InvocationFuture.get` / `LockSupport.park` wait is material to the steady-state timed path, not merely an empty stack label. The typical compute-path park is about 58 microseconds against a 165 microseconds per-operation quiet-day score, and the observed unpark path ties completion to the invocation response. It still does not prove the historical high-CV days are caused by those waits: this run is a 9.55% CV quiet day and `@OperationsPerInvocation(100)` smooths tails. Keep the production code unchanged. The next experiment should capture a noisy primary run and align its JMH sample windows with jmh-worker park sums, or use an explicitly diagnostic `OperationsPerInvocation(1)` run for one-to-one correlation. Preserve the same baseline, raw samples and environment record. #12081 remains related-only and cannot close this issue. -- 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]
