Rangsh commented on issue #12063: URL: https://github.com/apache/seatunnel/issues/12063#issuecomment-5579203914
Thanks @SEZ9 — status update on the evidence request. ### Fork Actions queue I had triggered a full diagnostics matrix on `Rangsh/seatunnel`, but those runs stayed **Queued for 36+ minutes** without starting (free fork runner backlog). I cancelled that pile-up so it would not block forever. I also do not have permission to dispatch `Benchmarks Diagnostics` on `apache/seatunnel` (403). If a maintainer can run it there with `pr_number=12173` plus a matching baseline ref, that remains the preferred same-runner comparison against the original report. ### Controlled local before / after (interim evidence) Same machine, same JDK 17, same JMH args (`-f 3 -wi 3 -i 5`), `-prof gc`, both methods and both `initialStoredJobCount` values: - **Before**: `origin/dev` (`8bea8c681`) jar - **After**: this PR branch (`991e67b40`) jar | Benchmark | Count | Before Score | Before Error% | Before CV% | After Score | After Error% | After CV% | | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | | `runningJobGrowth` | 0 | 186.425 us/op | 22.33% | 20.89% | 157.158 us/op | 33.87% | 31.68% | | `runningJobGrowth` | 1000 | 197.773 us/op | 14.26% | 13.34% | 124.843 us/op | 5.65% | 5.29% | | `completedJobHistoryGrowth` | 0 | 1146.134 us/op | 9.04% | 8.45% | 1041.015 us/op | 11.42% | 10.68% | | `completedJobHistoryGrowth` | 1000 | 1623.098 us/op | 54.38% | 50.87% | 839.794 us/op | 7.93% | 7.42% | GC secondary metrics (same runs): | Benchmark | Count | Before gc.time | After gc.time | Before alloc B/op | After alloc B/op | | --- | ---: | ---: | ---: | ---: | ---: | | `runningJobGrowth` | 0 | 38 ms | 22 ms | 375151 | 164414 | | `runningJobGrowth` | 1000 | 7 ms | 8 ms | 1845453 | 421489 | | `completedJobHistoryGrowth` | 0 | 65 ms | 40 ms | 2757946 | 1275670 | | `completedJobHistoryGrowth` | 1000 | 933 ms | 263 ms | 9677339 | 2705012 | Raw JMH JSON is on the local evidence workspace (`before-growth.json` / `after-growth.json`); I can attach artifacts to the PR if useful. ### Reading - Clearest variance win is at `initialStoredJobCount=1000` (especially `completedJobHistoryGrowth` CV 50.87% → 7.42%, with much lower gc.time / alloc). - `initialStoredJobCount=0` is mixed locally (`runningJobGrowth` CV still noisy). Absolute local scores are still **not** comparable to the GitHub Actions baseline in this issue. - This after jar includes both the fixture change and the single-`put` production change, so this table alone cannot attribute CV reduction to production vs fixture. ### Scope If you prefer, I can split the PR to land **fixture-only** first and move `storeFinishedPipelineMetrics` to a follow-up once same-runner attribution is clearer. PR description already documents durability-coverage and single-`put` correctness for points 3–4. Please advise whether to wait for an apache-side diagnostics run or proceed with a fixture-only split. -- 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]
