Monitor tasks and recover work

Retain the caller, application scope, exact request and key for any operation with effects. A job ID identifies native durable work. A task ID identifies its entry in the unified task view. Use the grant and scope required by the operation to read its status and evidence.

Choose a recovery action from observed state

1. Inspect a job or task using its exact ID. For a job, use read_v2_job. Locate a matching task by its source ID and page the relevant events or logs. The published sample job, final task event and fact readback agree on the receipt revision.
2. Reconcile a lost response before creating new work. Where the function supports exact replay, use the original caller, scope, typed request and key. The HTTPS sample recovers the same job and one publication.
3. Interrupt work only when the task reports interruptible and the interruption is authorized. Read the resulting state and whether committed effects remain. Interruption can stop further work while leaving earlier committed units in place.
4. Rerun eligible work only when rerunSupported and its scope and inputs are still valid. The export example returns a distinct task ID. Repeating that exact rerun returns the same new task.

For capacity diagnosis, use the scoped read_v2_capacity_metrics with the application Administrator grant. Keep capacity signals, such as queue age, separate from the evidence for one request's outcome.

Recovery prompt templates · Monitoring and recovery tools