[CI] Improve run-slow PR comments and remove CI recap (#49782)
* [CI] Improve run-slow PR comments and remove CI recap
1. run-slow comments (Nvidia + AMD):
- Always post a new comment per trigger instead of editing the old
one (prevents results from being buried in large PR threads)
- AMD reply_to_comment now attaches to the fresh comment created by
Nvidia for the same trigger (matches by post timestamp >= trigger
comment time)
- report jobs find the right comment by run URL instead of just the
header, so multiple triggers don't overwrite each other's results
- Shorten result comments: remove commit SHA table, add Grafana
dashboard link, wrap failure lists in <details>
2. Remove CI recap comment:
- update_pr_ci_dashboard_recap.py no longer posts the
<!-- ci-dashboard-recap --> comment after each PR CPU CI run
- Existing recap comments are cleaned up on the next CI run
- Grafana badge injection into the PR body is unchanged
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [CI] Strip dead code from update_pr_ci_dashboard_recap.py
Remove all Prometheus/Grafana querying and recap-rendering code that
became dead after the CI recap comment was removed. The script now
only injects the Grafana badge into the PR description and cleans up
any legacy recap/dashboard comments.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [CI] Update test file to match trimmed update_pr_ci_dashboard_recap.py
Remove tests for deleted functions (Prometheus querying, recap
rendering, quality_job_failed, recreate_ci_recap_comment). Replace
RecreateCiRecapCommentTest with DeleteRecapCommentsTest for the new
cleanup function. Update BadgeUpdateTest to not patch removed functions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [CI] Delete previous run-slow comment before posting new one
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [CI] Compact run-slow comment format per Tarek's suggestion
- Single-line format per hardware: **Nvidia/AMD** · status · ⚙️ [Workflow] · 📊 [Dashboard]
- Rename "Grafana" to "Dashboard" throughout
- Remove --- separator between Nvidia and AMD sections
- Failure details wrapped in <details><summary>Failure details</summary>
- Failure count is inline in the status line, not a separate header
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Remove unused urllib imports (ruff fix)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Simplify comment lookup: drop run_url matching in report jobs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Simplify AMD comment lookup: drop trigger_time coordination
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Remove legacy comment cleanup strings and stale complexity in report jobs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* report: delete+repost only when both sides have final results
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: move nv_still_running check to report job (not reply_to_comment)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>