Skip to main content

auto qa tuning

.github/auto-qa-tuning.json defines the initial Auto-QA policy for ACMM acmm:auto-qa-tuning. It is a declarative policy only: no workflow consumes it or changes CI settings automatically. A future tuner can use it to propose timeout updates for the plain and LUKS install jobs.

Use runs on main from the last 30 days that succeeded on the first try. Separate samples by workflow, job and runner. Exclude pull requests, reruns, failed, cancelled and skipped jobs so untrusted or incomplete runs cannot lower the budget. Keep the current value when fewer than 20 eligible samples exist.

For each target, compute the nearest-rank 95th percentile of job duration (excluding queue time), multiply by 1.5, and round up to five minutes. Clamp the result to 60–180 minutes, then limit movement to 10% of the current timeout per proposal. Apply the change limit last; if the round-up or the absolute bounds would need a larger change, keep the current value for maintainer review. Read the current timeout from the target workflow; do not duplicate it in this policy. Both jobs now allow 120 minutes.

Any proposed change must include the source run IDs, sample count, percentile, current timeout and proposed timeout in a reviewed PR. Maintainers should check for runner or workload changes before they accept it. Required checks, test assertions, security scans and workflow permissions are outside the tuning scope. Spikes in timeouts or failures need investigation, not an automatic relaxation of quality gates. Revert the workflow timeout change to roll back; this policy itself does not grant write permissions or enable a scheduled job.