Repository navigation
fix(ci): give Backend Lint 25 min so mypy is not cancelled mid-run - #9452
Conversation
The backend-lint job's floor on today's hosted runners is ~14-15 min: `check_black_formatting.py` runs `black --check` over the whole tree (7-8 min), flake8 takes ~2 min and mypy starts cold with under two minutes of the 15-min budget left. Observed today: - run 34211086807 (fix/cron-command-shell-brace-off): lint passed at 14m38s - run 34209554482 (PR #8947) attempt 1: cancelled at 15m18s, mypy running - run 34209554482 attempt 2: cancelled at 15m06s, mypy running 25 is the budget frontend-test and bundle-size already run on. No other change.
Opus 4.8 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
Design Review (Fable 5) — ✅ PASSDesign-level review of Design-Verdict: PASS A one-line, evidence-backed, fully reversible budget fix; the slower-alternative question is answered in the PR itself. Suggestions
[DESIGN-REVIEWED] 3996820 |
First Principles Review (Fable 5) — ✅ PASSPremise-level review of The change is fully verified against the repo: single-line timeout bump, both cited sibling jobs ( First-Principles-Verdict: PASS Verify the cited runs — What this change shipsIntent: stop the CI job budget from cancelling mypy on green backend-lint runs — a FIX. Inventory (1 item)
Checked: [FIRST-PRINCIPLES-REVIEWED] 3996820 |
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
iamwhatever
left a comment
There was a problem hiding this comment.
Tier 1 auto-approve: chore / build tooling (1 file). Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, SAST annotations clean, security checklist all-NO, AI reviewers green. Category: CI plumbing only -- raises the Backend Lint job timeout in .github/workflows/ci.yml so mypy is not cancelled mid-run; no runtime impact.
…irodotdev#9452) The backend-lint job's floor on today's hosted runners is ~14-15 min: `check_black_formatting.py` runs `black --check` over the whole tree (7-8 min), flake8 takes ~2 min and mypy starts cold with under two minutes of the 15-min budget left. Observed today: - run 34211086807 (fix/cron-command-shell-brace-off): lint passed at 14m38s - run 34209554482 (PR kirodotdev#8947) attempt 1: cancelled at 15m18s, mypy running - run 34209554482 attempt 2: cancelled at 15m06s, mypy running 25 is the budget frontend-test and bundle-size already run on. No other change.
Summary
One-line CI unblocker:
backend-lint(Backend Lint & Type Check) getstimeout-minutes: 25(was 15).The job's floor on today's hosted runners is ~14–15 min —
check_black_formatting.pyrunsblack --checkover the whole tree (7–8 min), the five gate scripts + isort take ~3 min, flake8 ~2 min, and mypy then starts cold with under two minutes of budget left — so a green PR gets its mypy step cancelled by the job timeout.Observed today (all
ubuntu-latest):34211086807(fix/cron-command-shell-brace-off)34209554482attempt 1 (PR #8947, frontend + docs, no Python)34209554482attempt 234209554482attempt 325 is the budget
frontend-testandbundle-sizealready run on in this file. Nothing else changes; the job's steps and gates are untouched.Why not make the steps faster instead
Black over the whole tree is the deliberate design of the shrinking-baseline gate (
scripts/check_black_formatting.py— it must see every file to detect a baselined file that has become clean). Speeding that up is a real change to the gate; this PR only stops the budget from cancelling a job whose steps all pass.Tests
Workflow-only change;
python3 -c "import yaml; yaml.safe_load(...)"parses. Verification is this PR's ownBackend Lint & Type Checkjob finishing.Pattern harvest
Rule candidate: a job whose observed floor is within ~10% of its
timeout-minutesis a flake generator — when a CI job's passing wall time approaches its budget (here 14m38s / 14m15s against 15), raise the budget or split the job before a green PR is cancelled, rather than rerunning. Checkable from the Actions API (jobcompletedAt - startedAtvs the workflow'stimeout-minutes).no linked issue: CI unblocker for an inherited red seen on PR #8947 (and any backend PR on a slow runner).