Rangsh commented on PR #12173: URL: https://github.com/apache/seatunnel/pull/12173#issuecomment-5612134815
Thanks @SEZ9 — confirming per item against current head `c0d59da` (also aligning with @DanielLeens' re-trace): 1. **Durability sample** — addressed before `c0d59da` in `ca82f95` / `bf1110b` / `3794d03`: first empty-pressure iteration + trial tear-down sampling restores WAL-failure visibility beyond “last batch of the fork only.” Mid-trial full `loadAll` under `initialStoredJobCount=1000` OOMs (seen on Diagnostics), so sampling stays scoped that way. The sample still reloads one representative key (same full-WAL cost as any key set) and asserts the rest of the batch remains resident; `c0d59da` only makes that coverage claim accurate in Javadoc/docs. Widening to a full-batch `loadAll` is a reasonable follow-up, not required to close this item for this PR. 2. **`storeFinishedPipelineMetrics` atomicity** — addressed in `ca82f95`: `get → merge → put` is wrapped in cluster-wide `finishedJobMetricsImap.lock(jobId)` / `unlock(jobId)` (`try`/`finally`). 3. **Docs** — addressed in `909f878`: `docs/en|zh/engines/zeta/benchmark.md` describe the growth methodology (unique keys; first empty-pressure + trial durability sample). Same-worker comparisons remain the recommended path in the Benchmarks workflow section (`Baseline → Candidate → Candidate → Baseline`); the fork before/after I posted earlier is observational only due to different hosts. 4. **Javadoc** — addressed in `ca82f95`: `storeFinishedPipelineMetrics` Javadoc states the lock + single TTL put contract (no “in memory” / historical `computeIfAbsent` rationale), with `@param` tags. 5. **Tests** — addressed in `ca82f95`: `JobHistoryServiceFinishedMetricsTest` asserts the observable single TTL `put` / merge result (no `never().computeIfAbsent`), plus null-metrics rejection and distinct-tag merge coverage, with explicit `lock`/`unlock` verification. `c0d59da` itself is only the durability-sample Javadoc precision fix. Happy to take a follow-up for full-batch `loadAll` if you want that separately. Please do the final pass when you have a chance. -- 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]
