Job Completion Semantics

A Job is the workload controller for work that is meant to run once and finish, rather than run forever, which is why the workload-choice section of the Pods module pointed here for finite tasks. A Job creates and owns Pods, and it keeps creating replacement Pods on failure until the number of Pods that exit successfully reaches the requested completions count - because of this "try again on failure" model, a Job's Pods must use restartPolicy: Never or OnFailure, never Always (which is meant for long-running workloads). Three settings shape how forgiving that retry behavior is: backoffLimit caps the total number of retries before the Job gives up and marks itself failed; activeDeadlineSeconds caps the total wall-clock time the Job is allowed to keep running at all, regardless of retry count; and ttlSecondsAfterFinished automatically cleans up the finished Job (and its Pods) after a delay, so completed one-off Jobs do not pile up forever. One design decision worth making deliberately up front: should a failed attempt be retried inside the same container process (an internal retry loop in your own code), or should Kubernetes retry it by creating a whole new Pod? Each has very different implications for logging, state, and how cleanly a partial failure can be resumed.

Analogy: Kubernetes is a thermostat, not a remote control. You declare the temperature; independent controllers measure reality and keep acting. Debugging means finding which sensor, rule or actuator prevents convergence.

Read the object as evidence

kubectl get RESOURCE NAME -n NAMESPACE -o yaml
kubectl describe RESOURCE NAME -n NAMESPACE
kubectl get events -n NAMESPACE --sort-by=.metadata.creationTimestamp
kubectl explain RESOURCE.spec

Do not memorize these as a ritual. The first command exposes desired and observed state, the second connects conditions and events, the third supplies a timeline, and the fourth checks the server's schema. Compare what the controller was asked to do with what it reports doing. Then test the narrowest hypothesis.

Scenario: A payment-processing batch job writes some, but not all, of its output rows to the database before crashing partway through. Kubernetes, doing exactly what it is supposed to do, creates a brand-new Pod to retry the whole Job - which reruns the entire batch from the beginning, including the rows that already succeeded and were already charged, silently double-charging some customers. Kubernetes' Job controller only guarantees that it will keep retrying until it sees a successful exit; it has no idea whether re-running your specific business logic from scratch is safe. That safety - called idempotency, meaning "running it twice produces the same result as running it once" - has to be designed into the application itself, often via a checkpoint or an idempotency key.

Define success and retry ownership

A Job considers successful Pods, not business transactions. Set backoffLimit, activeDeadlineSeconds and restartPolicy intentionally, and make the command return nonzero on real failure. With OnFailure, retries occur inside a Pod and logs may be harder to separate; Never creates new Pods whose attempts remain visible. Use kubectl describe job and inspect every owned Pod.

Scenario: A script catches an exception, logs it and exits zero. Kubernetes marks the Job complete although no records were processed. Exit status is part of the workload contract.
Warning: Retrying an externally visible operation requires an idempotency key or transactional checkpoint. backoffLimit reduces infinite retries; it does not create exactly-once semantics.

Production reasoning

Ask four questions: Who owns this object? What dependency must become ready next? Which controller reports the blocking condition? What evidence would disprove my current theory? This prevents symptom-driven changes. Record the context, namespace, object generation, image digest and recent rollout before mutation; a recreated Pod may erase the evidence you needed.

Warning: Running is not the same as ready, healthy, durable or correct. Kubernetes status is layered. Confirm the application-level outcome as well as the object state.
Goal: Put this model into practice in the Kubernetes lab job-restart-policy. Open /labs/kubernetes, choose job-restart-policy, predict the failure path before changing anything, then use the simulator's check command to validate the finished state.

Deliberate practice

Before the lab, write the expected object relationship and the first three commands you will run. Afterward, explain why the fix converged and name one tempting change that would only mask the symptom. Repeat using an explicit namespace and a structured output format. This prediction-observation-explanation loop is what turns command familiarity into production judgment.

45-minute investigation

  1. Map (5 min): draw the owner-to-child chain and mark every namespace, selector, identity and dependency involved.
  2. Predict (5 min): write one expected status condition, one likely event and one log or metric signal before opening the lab.
  3. Observe (10 min): collect YAML, describe output and ordered events. Do not mutate state. Record which observation disproves your first theory.
  4. Repair (15 min): make the smallest declarative correction, watch the responsible controller converge, and verify the user-facing path rather than stopping at Running.
  5. Stress (10 min): change one relevant constraint - replica count, label, readiness, resource value or placement rule - predict the outcome, observe it, then restore the known-good declaration.
Tip: Keep a short incident note with symptom, evidence, hypothesis, change, result. Across four sections this produces a reusable runbook instead of a pile of remembered commands.