AutoCoder Engine PR validation remained Waiting to run although all three rust-large runner pods were Ready. Runner processes and Docker were alive, Gitea health checks passed, but runner logs showed no accepted tasks after 09:22 UTC.
Restarted the three repository-managed deployments:
gitea-act-runner-rust-1gitea-act-runner-rust-2gitea-act-runner-rust-3All replacements became 3/3 Ready with zero restarts. The queued validation was accepted immediately and completed successfully in 6m59s.
At 19:02 UTC, another validation initially appeared as Waiting to run in commit status. The Rust runners were restarted at 19:04 UTC on the assumption that the polling fault had recurred.
Direct inspection of Gitea action task 9095 and the runner's Docker containers then proved that this workflow used ubuntu-latest, not rust-large, and had already been running on the general runner since 19:02 UTC. The commit-status description was stale. The second Rust-runner restart was unnecessary and is not evidence of recurrence.
Pod readiness alone is insufficient runner-health evidence, but a queued-looking commit status alone is also insufficient failure evidence. Before recovery, correlate the workflow's requested label, Gitea action-task state and update time, registered runner busy state, and the task container on the assigned runner. Alert only when an eligible job is genuinely unassigned or has stopped progressing for a bounded interval.