mirror of
https://github.com/actions-runner-controller/actions-runner-controller.git
synced 2026-09-30 18:46:48 +02:00
Nikola JokicandCopilot App
e2768dddbf
When the runners reject the runner spec they were given, the listener has to stop acquiring jobs the scale set cannot run. The controller removed the listener but then deleted the EphemeralRunnerSet, and left the AutoscalingRunnerSet phase on Running. AutoscalingRunnerSetPhaseOutdated was declared and read, but never assigned by anything. Nothing held the scale set switched off as a result. The next reconcile saw a missing EphemeralRunnerSet, created it, created a listener for it, and the fresh runners rejected the same spec again, so the scale set churned through create and teardown cycles against the Actions service instead of resting. Record the outdated phase and keep the EphemeralRunnerSet, pinned to zero replicas and patch id. It releases every runner that is not executing a job while the phase keeps the listener from being rebuilt, and the revision bookkeeping that decides when the scale set may run again is preserved. Recovery is driven by the spec update that the phase is waiting for: it moves the phase back to pending, and the runner spec is then republished to the set with an advanced revision even when the runner spec itself did not change. The revision is what tells the EphemeralRunnerSet to stop judging itself by the runners that failed, so without advancing it a scale set could only be recovered by editing the pod template, and an edit to anything else would switch the listener back on against a set parked at zero. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>