Pause, resume, restart or stop safely
Choose a service-control action by its consequences, not by which status badge you want to clear. These commands change running work. You need operator permission for the exact controller-launched service and the CLI on PATH; a recorded agent request is not permission by itself.
Review the target
Inspect fleetctl show "SERVICE_ID", substituting the exact identity. Check owner, purpose, working directory, recorded command, PID/process group, lease and recent events. Preserve output or unsaved work before a disruptive action. Claimed and unmanaged processes are not eligible for these controller controls.
Choose one action
| Intent | Command with the inspected identity | Consequence |
|---|---|---|
| Suspend scheduling | fleetctl pause "SERVICE_ID" | Retains process group, memory and ports |
| Continue paused work | fleetctl resume "SERVICE_ID" | Work and expiry cleanup can resume |
| Start the recorded command again | fleetctl restart "SERVICE_ID" | Stops the old group, waits, then replays recorded argv |
| End the service | fleetctl stop "SERVICE_ID" | Requests SIGTERM, then escalation after the grace period |
Do not run the table as a sequence. A restart is appropriate only if the recorded command is still correct for the project. It is not a substitute for reviewing configuration changes. kill is immediate SIGKILL with no grace period and is not the normal stop path.
Verify the action, then the task
Inspect the same record and its events after the action. On resume or restart, confirm a fresh health-check result if configured and separately check the task's useful result. On stop, do not treat an accepted request as proof all work disappeared. Escaped descendants or a surviving listener require their own current identity review.
If identity verification refuses an action, do not bypass it with a raw PID signal. Re-read current observations and consult the owner. The CLI control reference describes verify-before-signal behavior.