You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Maintained by the CI profiler routine, which runs every 4h. Each run replaces this body and carries the history rows over.
Headline (run 2026-10-12 04:17 UTC)
Window: 2026-10-11 04:17 → 2026-10-12 04:17. 23 PRs merged through the merge queue. Steady state: merge-queue and PR latency match the last profile. No new finding clears the filing bar. One human-filed issue, #1414, has a PR open (#1415). Fresh numbers are posted on it.
Sample: run counts cover the full 24h: 379 runs (260 pull_request, 92 merge_group across 4 workflows, 23 push, 1 schedule, 3 workflow_dispatch). Run-level detail covers the 300 runs listed (3-page cap), which reach back to 15:38. Job detail covers 85 runs: all 22 CI merge_group runs listed, all 17 push runs, all 3 dispatch runs, the nightly schedule run, 34 of 101 CI pull_request runs (every 3rd), and 1–2 runs each of fail-fast, janitor, Audit and Pin check. Also 15 merged-PR timelines and the ruleset. 0 job logs. About 120 REST requests. GraphQL is blocked here.
metric
24h
PRs merged
23 (23 CI merge_group runs/day; 22 sampled, all success; no re-runs)
Merge queue enqueue → merge
p50 8.0 / p90 8.3 / max 8.7 min (n=15)
Evictions
0 in 15 timelines
Bursts
Largest was 3 entries (02:49 and 23:48), below max_entries_to_build 5.
merge_group created → ci-ok
After #1386: p50 7.3 / p90 7.9 (n=19, range 7.07–7.90). All 22: p50 7.33 / p90 7.98 / max 9.87 (the max is from 2 pre-#1386 runs at 17:50 that ended on e2e-build-windows). Last job before ci-ok: coverage 16, Gradle e2e leg 4, e2e-build-windows 2 (pre-#1386).
PR run created → ci-ok
Full runs: p50 7.45 / p90 8.9 / max 9.2 (n=12). The last job was coverage in 11 of 12. 22 of 34 sampled PR runs were drafts (only ci-ok).
push to main created → ci-ok
p50 3.43 / p90 3.82 (n=17). An e2e-full leg was last in 10 of 17. 1 failure: cargo stable lock-v1 toolchain install at 22:58, already noted last run.
Linux queue wait p50 / p90 / max
Depot 0.02 / 0.02 / 0.18 min (n=2,645)
macOS queue wait p50 / p90 / max
0.12 / 0.15 / 0.25 (n=91, nightly and dispatch only). No macOS job ran on PR, merge_group or push.
Windows queue wait p50 / p90 / max
0.03 / 0.05 / 0.07 (n=89)
Job-min/day (Linux / Windows / macOS)
~4.0k / 0.33k / 0.19k. By event: mg 30.8 L + 4.4 W per run (708 + 101/day); CI PR 11.4 per run averaged over drafts (~139 runs/day → ~1.6k); push 29.1 per run (~580/day, of which full tier 14.2); nightly 298 / 67 / 54; workflow_dispatch 3 full-matrix runs by a maintainer validating janitor branches (ci-janitor/go-work-symlinked-tmp ×2, agent/fix-ci-windows-pnpm-shims), ~700 / 162 / 132 in total; Merge queue fail-fast 7.6 hosted-Linux min per mg run (~175/day). The rise from ~2.7k comes from the dispatches and from PR volume (~139 CI PR runs vs 104). Per-run cost is unchanged.
Critical path (unchanged): clippy is 1.3 min (clippy 21 s, then cargo check --all-targets 41 s, p90 45 s). coverage has needs: clippy and runs 5.4 min, of which "Run tests with coverage" is p50 323 s / p90 357 s (n=22).
Watched, not filed (below the bar or not enough evidence):
Shard coverage into 2 partitions: at most ~1.3–1.8 min off every merge and full PR run. Still unconfirmed, because the log tool can't reach the compile/CLI-test part of the step.
clippy → coverage serialization: ~1.3 min per merge and per full PR run. coverage reads clippy's reuse/pr_unchanged outputs, so a fix would move those outputs into a cheaper job. Not filed: the saving is under the 2-min bar and the change is M.
coverage "Warm the coverage cache" on push: ~26 Depot min/day, below the bar.
Merge queue fail-fast: ~175 hosted-Linux min/day, off the critical path, and free on this public repo.
ROI on one scale: (thousands of job-min/day weighted Linux ×1, Windows ×2, macOS ×3, plus 1 per critical-path or PR-latency minute) × confidence ÷ effort (S=1, M=2, L=3).
~7.3 min for each entry past the 5th in a burst of more than 5. No such burst in 72h; the largest this window was 3.
human decision (commented 10-12 00:20, no new comment today)
~0.5
#1414 full tier (e2e-full, yarn-berry-full, cargo-vex-matrix-full) runs on every push to main
~285 Depot job-min/day (~580 rounded per job). 29 of 50 push jobs, 14.2 of 29.1 job-min per push run.
filed by a human 10-12 03:41. PR #1415 is open, merge state blocked. Fresh numbers.
Closed since the last profile: none. #1173 (test windows/macos redundant build) was closed not planned by a human on 10-11 19:12, together with PR #1396. It will not be refiled.
Open ci-perf / ci-janitor PRs:#1415 (fixes #1414).
[agent] There are now two open dashboard issues: this one and #1175, both titled "CI performance dashboard" and labeled ci-perf-dashboard. The body says it is maintained in place, so the profiler routine should update one issue rather than open a new one each run. I'm not closing either one, because I can't tell which issue the routine will update next. Whoever owns the profiler should close the copy it no longer updates.
Maintained by the CI profiler routine, which runs every 4h. Each run replaces this body and carries the history rows over.
Headline (run 2026-10-12 04:17 UTC)
Window: 2026-10-11 04:17 → 2026-10-12 04:17. 23 PRs merged through the merge queue. Steady state: merge-queue and PR latency match the last profile. No new finding clears the filing bar. One human-filed issue, #1414, has a PR open (#1415). Fresh numbers are posted on it.
Sample: run counts cover the full 24h: 379 runs (260 pull_request, 92 merge_group across 4 workflows, 23 push, 1 schedule, 3 workflow_dispatch). Run-level detail covers the 300 runs listed (3-page cap), which reach back to 15:38. Job detail covers 85 runs: all 22 CI merge_group runs listed, all 17 push runs, all 3 dispatch runs, the nightly schedule run, 34 of 101 CI pull_request runs (every 3rd), and 1–2 runs each of fail-fast, janitor, Audit and Pin check. Also 15 merged-PR timelines and the ruleset. 0 job logs. About 120 REST requests. GraphQL is blocked here.
max_entries_to_build5.ci-oke2e-build-windows). Last job beforeci-ok:coverage16, Gradle e2e leg 4,e2e-build-windows2 (pre-#1386).ci-okcoveragein 11 of 12. 22 of 34 sampled PR runs were drafts (onlyci-ok).ci-oke2e-fullleg was last in 10 of 17. 1 failure:cargo stable lock-v1toolchain install at 22:58, already noted last run.workflow_dispatch3 full-matrix runs by a maintainer validating janitor branches (ci-janitor/go-work-symlinked-tmp×2,agent/fix-ci-windows-pnpm-shims), ~700 / 162 / 132 in total;Merge queue fail-fast7.6 hosted-Linux min per mg run (~175/day). The rise from ~2.7k comes from the dispatches and from PR volume (~139 CI PR runs vs 104). Per-run cost is unchanged.Critical path (unchanged):
clippyis 1.3 min (clippy 21 s, thencargo check --all-targets41 s, p90 45 s).coveragehasneeds: clippyand runs 5.4 min, of which "Run tests with coverage" is p50 323 s / p90 357 s (n=22).Watched, not filed (below the bar or not enough evidence):
coverageinto 2 partitions: at most ~1.3–1.8 min off every merge and full PR run. Still unconfirmed, because the log tool can't reach the compile/CLI-test part of the step.clippy→coverageserialization: ~1.3 min per merge and per full PR run.coveragereadsclippy'sreuse/pr_unchangedoutputs, so a fix would move those outputs into a cheaper job. Not filed: the saving is under the 2-min bar and the change is M.coverage"Warm the coverage cache" on push: ~26 Depot min/day, below the bar.Merge queue fail-fast: ~175 hosted-Linux min/day, off the critical path, and free on this public repo.pr_unchanged) is still not observable: 0 of 34 sampled PR runs took that path. Across three profiles, 116 PR runs have been either full 29-job runs or 1-job drafts.Merge-queue settings (ruleset 24668460, read 04:2x): unchanged.
max_entries_to_build5,max_entries_to_merge5,min_entries_to_merge_wait_minutes5, ALLGREEN, timeout 90.History
Open
ci-perfissues by ROIROI on one scale: (thousands of job-min/day weighted Linux ×1, Windows ×2, macOS ×3, plus 1 per critical-path or PR-latency minute) × confidence ÷ effort (S=1, M=2, L=3).
e2e-full,yarn-berry-full,cargo-vex-matrix-full) runs on every push to mainblocked. Fresh numbers.Closed since the last profile: none. #1173 (test windows/macos redundant build) was closed not planned by a human on 10-11 19:12, together with PR #1396. It will not be refiled.
Open
ci-perf/ci-janitorPRs: #1415 (fixes #1414).Merged ci-perf / ci-janitor PRs (last 7 days) and measured effect
ci-ok9.65 → 7.3 min (n=7 → 19, p90 7.9). Enqueue → merge 10.3 → 8.0. Windows min per mg run 8.2 → 4.4. Delivered.pr_unchangedpath.ci-ok21.8 → 7.45. Delivered.cargo stable lock-v1). Re-passed on the next push.Method notes
filter=all.started_at−created_at. Skipped jobs are excluded.created_at→ci-okcompleted_at.total_countper event.get_job_logsreturns at most the last 5,000 lines, even with a largertail_lines.Generated by Claude Code