SEZ9 commented on issue #11549: URL: https://github.com/apache/seatunnel/issues/11549#issuecomment-5140617804
Closing this discussion — the direction proposed here was accepted and everything has landed: - **#11553 (merged)**: the benchmark harness lives in the main repo under `seatunnel-cli/benchmark/` (option a), versioned with the CLI it measures. Review hardening along the way: skipped requested gates are excluded from every pass metric (`full_gate_passed`), per-task fault isolation, env-var-first credential guidance, PostgreSQL port mapping per backend, and wrapper robustness fixes. - **#11551 + #11605**: website documentation (en/zh) — methodology, 7-model measured results, model selection guide, and a "run it yourself" guide. - Defects surfaced by benchmark runs while this discussion was open: #11508 (fixed in #11509, merged), #11510 and #11511 (upstream, still open), plus the bedrock-mantle provider gap (#11547 → #11548, merged). Follow-ups tracked separately rather than in this discussion: - CI smoke subset (10–15 L1-only tasks gating CLI PRs at zero API cost) — worth its own proposal once the harness has been exercised by more contributors. - Incremental semantic `verify` probes for the remaining executable tasks. - Data-driven CLI accuracy improvements from the 13 all-model-failure clusters (Doris/StarRocks metadata, conditional-routing golden example, PG-CDC prerequisites, wide-DAG wiring) — to be A/B measured with this benchmark. Thanks @yzeng1618 @davidzollo @DanielLeens for the reviews that materially hardened the metric contract. -- 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]
