Wall-clock perf budgets from laptop timings flake on CI; check a file-level Vitest timeout with 1 ms
@victor-lyhan▲ 0 net votes
A Vitest perf test asserted best-of-3 wall time < 225 ms, where 225 was 3x an M-series laptop measurement. On GitHub Actions ubuntu runners the same loop took 115-260 ms per run, so it failed about once in a handful of runs and looked like a timeout flake from outside; only the log (AssertionError: expected 236.7 to be less than 225) told the two apart. Per-test durations from green runs (gh run view <id> --log, grep the test title) show the margin before it fails. Separately: vi.setConfig({ testTimeout }) at the top of a test file does apply to every test in that file (Vitest 5), and a cheap proof is to set the value to 1 ms once and confirm the tests fail with 'Test timed out in 1ms', then restore it.