gha-doctor — psf/requests F · 27/100

Workflow checkup (8 files)

Run history: psf/requests (last 120 runs)

Flaky jobs (failed and passed on the same commit):

Flaky tests (seen failing in 4 of 4 flaky-failure logs; the same commit also passed):

Matrix balance (a matrix finishes when its slowest shard does; billable minutes unchanged — this is PR feedback latency):

Worst: build — slowest shard (3.15-dev, windows-latest) 4.5m vs fastest (3.12, ubuntu-22.04) 1.6m (median of 12 runs).

Wasted compute: 1055 of 4490 minutes (failed runs 9 + retries 1045).

Estimated cost (GitHub-hosted rates; free for public repos on standard runners): $44.93 for the sample (5616 billable min), of which $10.60 went to failures/retries and $9.01 to per-job minute round-up.

Cache: 0 caches, 0 MB (0% of the 10 GB limit); 0 MB stale (unused 7+ days), 0 MB pinned to PR refs.

Artifacts: 18 total; 0 MB not yet expired in the 18 most recent. Top producer: dist (10 uploads, 2 MB, kept 5d).

Cache hit rate (12 sampled job logs): 100% — 1 restores: 1 hits, 0 partial, 0 misses.

Top wins

1. Cut failures and retries — ~$14.18/mo — 9 failed-run min + 1045 retried-attempt min bought nothing ($10.60 of the sampled spend); worst flake: "build (3.12, windows-latest)" in "Tests"; most-seen flaky test: "tests/test_requests.py::TestRequests::test_pyopenssl_redirect" (pytest, failed in 4 sampled logs)
2. Consolidate tiny jobs — ~$12.05/mo — per-job round-up billed 1126 min the jobs never used (20% of spend) — merge short jobs or shrink the matrix
3. Stop double-running PR pushes — 3 workflows trigger on both unscoped push and pull_request — every PR commit runs twice; scope push to your default branch (see: gha-doctor --explain D013)
4. Cache dependencies — 5 setup steps download the same dependencies every run (gha-doctor --fix handles this (D003))
5. Cancel superseded runs — 5 workflows have no concurrency group — runs for already-replaced commits keep burning minutes (gha-doctor --fix handles this (D001))

dollar wins projected to 30 days from the 22.4-day run sample; private-repo rates, public repos run free.

Health score: F (27/100)

Basis: static checks + run history.