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
{{ message }}
Repository navigation
CI perf: full tier (41 legs) runs on every push to main, not just nightly #1414
The ci.yml header says the full tier is nightly, but e2e-full, yarn-berry-full and cargo-vex-matrix-full also run on every push to main: 41 legs per landing, most done in under a minute, which came to 1,845 of the 2,330 push-run jobs across 45 pushes on Oct 10-11. Gating them like e2e-docker (schedule, workflow_dispatch and the v5 integration PR) keeps the full tier on the nightly and release-qa dispatches; the trade-off is that a regression only that tier catches shows up at the next nightly instead of on the landing. PRs here are collaborators-only, so could a maintainer cherry-pick this signed commit (it also updates the two scheduling tests that pinned the old condition, and scripts/tests pass) or grant PR access? main...austenstone:socket-patch:ci-full-tier-nightly
[agent] Triage: confirmed on origin/main 1779317. .github/workflows/ci.yml:2032 (e2e-full), :2286 (yarn-berry-full) and :2426 (cargo-vex-matrix-full) only exclude pull_request and merge_group, so they run on every push to main. Push run 38197830089: 41 of its 50 non-skipped jobs are these legs (~21 runner-minutes wall, ~41 billed job-minutes). The nightly schedule and workflow_dispatch already cover this tier.
[agent] Claiming this issue (shared root cause: full-tier jobs gated on 'not PR/queue' instead of nightly/dispatch). Branch: agent/fix-ci-full-tier-nightly. Claim-ID: 2026-10-12T03:50:35Z-8f4246
CI profiler — fresh numbers (window 2026-10-11 15:38 → 2026-10-12 04:00 UTC, 17 push-to-main CI runs, all jobs fetched with filter=all)
per push run
per day (~20 CI pushes/24h)
Non-skipped jobs
50.0
~1,000
of which e2e-full* / yarn-berry-full / cargo-vex-matrix-full
29.0
~580
Full-tier wall job-min (Depot Linux)
14.2
~285
Full-tier job-min rounded up per job
29.1
~580
All other push jobs (coverage 1.9, hosted-e2e 1.4, clippy 1.3, yarn-berry 1.3 ×2, e2e-build 1.1, …)
14.8
~300
Full-tier legs are short: p50 0.47 min, p90 0.75 min, so most of their cost is per-job setup.
push created → ci-ok: p50 3.43 / p90 3.82 min. An e2e-full leg is the last job before ci-ok in 10 of 17 runs, so gating it should also take a little off the push verdict (unmeasured; coverage would become last).
Merge queue and PR runs are unaffected: no full-tier job ran on any of the 22 merge_group or 12 full PR runs sampled.
Expected effect of CI: run the full e2e tier nightly, not per push (#1414) #1415: about half of push-to-main job time, ~285 Depot job-min/day (~580 rounded per job). Nightly and release-qa dispatches keep the tier, so coverage stays at nightly.
Activity
mikolalysenko commentedon Oct 12, 2026
[agent] Triage: confirmed on origin/main 1779317.
.github/workflows/ci.yml:2032(e2e-full),:2286(yarn-berry-full) and:2426(cargo-vex-matrix-full) only exclude pull_request and merge_group, so they run on every push to main. Push run 38197830089: 41 of its 50 non-skipped jobs are these legs (~21 runner-minutes wall, ~41 billed job-minutes). The nightly schedule and workflow_dispatch already cover this tier.[agent] Claiming this issue (shared root cause: full-tier jobs gated on 'not PR/queue' instead of nightly/dispatch). Branch: agent/fix-ci-full-tier-nightly. Claim-ID: 2026-10-12T03:50:35Z-8f4246
Generated by Claude Code
mikolalysenko commentedon Oct 12, 2026
[agent] Fix in #1415.
Generated by Claude Code
CI: skip e2e binary upload on lean pushes
mikolalysenko commentedon Oct 12, 2026
CI profiler — fresh numbers (window 2026-10-11 15:38 → 2026-10-12 04:00 UTC, 17 push-to-main CI runs, all jobs fetched with
filter=all)e2e-full*/yarn-berry-full/cargo-vex-matrix-fullci-ok: p50 3.43 / p90 3.82 min. Ane2e-fullleg is the last job beforeci-okin 10 of 17 runs, so gating it should also take a little off the push verdict (unmeasured;coveragewould become last).release-qadispatches keep the tier, so coverage stays at nightly.Example runs: 38197830089, 38187527055, 38186163932. The profiler will report the before/after on the dashboard (#1182) once #1415 merges.
Generated by Claude Code