ci-queue-watch.yml

name: CI queue watch

# Scheduled sentinel for #2499's "multi-hour head-of-line blocks page
# instead of silently parking green PRs" option. 2026-08-27 the whole
# quality pipeline queued for hours with nothing running (54 runs parked
# while ~5 jobs were in flight); the people waiting found out by polling.
# This sweep turns that observable into a labeled tracking issue, and
# closes it with recovery evidence when the queue drains again.
#
# Cron beats long-lived watchers here: a stuck queue cannot starve the
# alarm into existence the way it starves everything else, because a
# scheduled dispatch only needs one scheduling slot every 15 minutes —
# and during the exact degradation it watches, even a late sweep still
# fires (runs stay queued), which is the point.
#
# Deliberately NOT part of quality.yml: an in-pipeline alarm would wait
# behind the very router dependency that stalls. Runs the repo's own
# scripts/ci-queue-watch.py so the detection logic is inspectable and
# dry-runnable outside Actions (--dry-run validated against live queue
# state for #2499: 18 runs parked beyond 60m detected mid-stall).

on:
  schedule:
    - cron: "*/15 * * * *"
  workflow_dispatch:

# #3313: job-scoped, workflow level defaults to none -- see the same note in
# backup-staleness-watch.yml.
permissions: {}

concurrency:
  group: ci-queue-watch
  cancel-in-progress: false

jobs:
  sweep:
    runs-on: ubuntu-latest
    timeout-minutes: 5
    permissions:
      # Reading the run queue to find stalled runs ...
      actions: read
      # ... and writing the labeled tracking issue the finding becomes.
      issues: write
    steps:
      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
        with:
          persist-credentials: false
      - name: Sweep for runs parked in the queue beyond the stall threshold
        env:
          GH_TOKEN: ${{ github.token }}
        run: python3 scripts/ci-queue-watch.py