Lifecycle controls

Lifecycle commands operate through durable state and revalidated native process identity. They never target a PID solely because it was recorded in the past.

Inspect before changing

$ jobman status JOB
$ jobman show JOB
$ jobman list --active

Use JSON output when a script must decide which operation is valid. Repeating a completed or conflicting lifecycle operation returns a stable conflict error rather than silently changing unrelated state.

Cancel

$ jobman cancel JOB
$ jobman cancel job JOB
$ jobman cancel run JOB -1

cancel JOB and cancel job JOB are equivalent. Cancellation prevents future runs. In v1, canceling the selected active run also cancels its owning job, so a retry does not follow.

The supervisor records cancellation intent, requests a graceful tree-wide stop, waits for the job’s grace period, and optionally forces termination. Native signal details differ; see Platform support.

Pause and resume

$ jobman pause JOB
$ jobman resume JOB

Pause/resume is a best-effort platform feature implemented for the managed process tree. Pause duration does not consume run or whole-job timeout budget. The target might still interact with external systems immediately before the native suspension takes effect, so pause is not a transactional application checkpoint.

Queued or prerequisite-waiting work can also be paused. Resume returns it to its prior phase.

Wait without taking ownership

$ jobman wait JOB

wait blocks the client until the durable job is terminal. It does not own the target and closing the waiting client does not cancel the job. The command’s result reflects the managed job outcome according to the CLI contract.

Rerun immutably

$ jobman rerun JOB
$ jobman rerun --name replacement JOB
$ jobman run --rerun JOB --wait

Rerun submits a new job using the prior effective immutable specification. It does not mutate or reopen the original history. Policy flags are rejected when cloning through run --rerun; only documented submission controls such as the new name and waiting behavior may differ.

Recover uncertain state

If a supervisor or host stopped unexpectedly, inspect first:

$ jobman doctor --json
$ jobman show --json JOB

Use doctor --repair only after reviewing the report. Recovery checkpoints the WAL, reconciles provably stale ownership, and wakes due notification work; it does not invent a successful target result.