Rangsh commented on issue #12063: URL: https://github.com/apache/seatunnel/issues/12063#issuecomment-5578672888
Thanks @SEZ9 — agreed on keeping the evidence bar high and the scope tight. I updated [#12173](https://github.com/apache/seatunnel/pull/12173) to answer points 3 and 4 in the PR description: 1. **Durability coverage after moving reload to trial**: write-through already fails the measured `put`/`delete` if the WAL append fails; iteration tear-down still does resident cardinality/content checks; trial tear-down still runs one `evict` + `loadAll` sample for the last batch. So we are removing full-WAL replay from the measured-adjacent path, not removing durable coverage. 2. **Single TTL `put` state/concurrency**: new-job / existing-job final stored values match the old merge semantics; only the redundant empty write-through is removed. The only production caller (`JobMaster.savePipelineMetricsToHistory`) already serializes per job via `metricsLock`. Unit tests cover both cases in `JobHistoryServiceFinishedMetricsTest`. On points 1–2 (JMH JSON + GC/JFR, same-runner before/after including `initialStoredJobCount=1000`): - I do not have permission to dispatch `Benchmarks Diagnostics` on `apache/seatunnel` (403). - I have started the same workflow matrix on my fork for `dev` (before) and `improve/zeta-job-lifecycle-growth-stability-12063` (after), both methods, `initialStoredJobCount=0/1000`, JDK 11, GC + JFR: - https://github.com/Rangsh/seatunnel/actions/workflows/benchmarks_diagnostics.yml - If a maintainer can also run `Benchmarks Diagnostics` on `apache/seatunnel` with `pr_number=12173` (and a matching baseline ref from the original report), that will give the preferred same-runner comparison against the issue baseline. On scope: if the same-runner data shows CV reduction is dominated by the fixture change, I am happy to **land the fixture stabilization first** and move the `storeFinishedPipelineMetrics` single-write change to a follow-up PR. I will post the fork diagnostics artifacts / Score-Error-CV table here as soon as the runs finish. -- 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]
