Operations
Runner health and recovery
Understand heartbeats, reconnects, leases, retries, and retained work.
Runner health
The Runner page reports registration, heartbeat recency, protocol and CLI versions, active jobs, capacity, and enabled repositories. Online means a registered instance has a recent heartbeat.
Outbound reconnects
The runner reconnects with capped backoff when its WebSocket drops. Active work is tracked by leases so a short connection interruption does not immediately duplicate a job.
Safe retry and recovery
If a runner cannot report back, an expired lease returns the same work to the queue for another attempt. Issue Planner retains its interview and provider settings. Pull-request implementation uses preserved workspaces where required.
Frequently asked questions
Will a socket reconnect duplicate active work?
Short reconnects preserve active work. Leases and artifact verification protect later retries.
What should I do when the runner is offline?
Confirm the machine and Docker are running, restart orchestrator run, and inspect token validation and network errors.
Does Issue Planner lose the interview?
No. The plan and conversation are retained, and the same pinned provider settings resume after recovery.
When is work returned to the queue?
When the active lease expires without a successful renewal or completion report.