Job Run State

Summary

ops.job_run is the durable record of queued, running, completed, and failed job execution.

Problem

Automation that only exists in process memory cannot be inspected, retried, or explained after the process exits. Operators need to know what ran, why it ran, who requested it, and what happened.

Principle

Persist job execution state as platform data. Runtime transport can change, but the run record remains the audit and control point.

Platform Shape

Job run persistence and read models live under services/ops/execution_state/jobs. The shared runner under services/execution/jobs/runner records standardized results, errors, timing, and payloads. The execution modes doc describes how both database and Cloud Tasks dispatch still write durable run state.

Source Evidence

  • docs/reference/job-execution-modes.md
  • services/ops/execution_state/jobs
  • services/execution/jobs/runner
  • docs/engineering/concepts/run-artifacts-and-observability.md
  • docs/engineering/architecture/worker-leases-and-recovery.md