Repository navigation

CI queue watch #202

	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:

	permissions:
	actions: read
	issues: write

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

	jobs:
	sweep:
	runs-on: ubuntu-latest
	timeout-minutes: 5
	steps:
	- uses: actions/checkout@v7
	- name: Sweep for runs parked in the queue beyond the stall threshold
	env:
	GH_TOKEN: ${{ github.token }}
	run: python3 scripts/ci-queue-watch.py