SEZ9 commented on issue #12058: URL: https://github.com/apache/seatunnel/issues/12058#issuecomment-5578076002
@Rangsh Thanks for the clarification and the correction — that summary matches my understanding of this issue. To confirm the two points you raised: 1. Yes, this issue is an investigation task first. The goal is to explain the within-run variance on `checkpointIdAtomicIncrement` / `checkpointOverviewIncrementalUpdate` with reproducible evidence, and a PR is not a prerequisite for making progress here. 2. Yes, #12081 should be framed as **Related to #12058** rather than `Fixes`, and merging it should not close this issue. Treating the sync-collapse as a working hypothesis (not a confirmed root cause) and the call-count tests as method-call-count parity only is the right framing. I'll keep #12058 open regardless of what happens with #12081. Your proposed next step is exactly what I was hoping for: one method (`checkpointOverviewIncrementalUpdate`), existing measurement settings retained, on a consistent local machine. When you post, the following would be most useful: - Score / Error / CV for the reproduction, plus the JDK, machine, JMH settings, and state-store config used so others can reproduce. - Per-iteration observations that distinguish the normal iterations from the slower ones, and what runtime behavior they line up with. - Hypotheses you have ruled in or out, including the `file:///` LocalFileSystem `hsync`/`hflush` semantics vs real HDFS comparison you mentioned. - Any open questions where you are blocked or unsure how to isolate a factor. The existing fork-Actions evidence from #12081 is welcome as input; just flag it as such rather than as closing evidence, as you already suggested. Once we have the local reproduction and the initial observations here, we can decide together whether a benchmark correction, a production change, or additional diagnostics is the right next move. <!-- streview-comment:878 --> -- 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]
