A job, in contrast to a deployment (or a replication controller), runs a pod with any number of replicas to completion. A job tracks the overall progress of a task and updates its status with information about active, succeeded, and failed pods. Deleting a job will clean up any pod replicas it created. Jobs are part of the Kubernetes API, which can be managed with oc commands like other object types.

See the Kubernetes documentation for more information about jobs.

Creating a Job

A job configuration consists of the following key parts:

  • A pod template, which describes the application the pod will create.

  • An optional parallelism parameter, which specifies how many pod replicas running in parallel should execute a job. If not specified, this defaults to the value in the completions parameter.

  • An optional completions parameter, specifying how many successful pod completions are needed to finish a job. If not specified, this value defaults to one.

The following is an example of a job resource:

apiVersion: batch/v1
kind: Job
  name: pi
  parallelism: 1    (1)
  completions: 1    (2)
  template:         (3)
      name: pi
      - name: pi
        image: perl
        command: ["perl",  "-Mbignum=bpi", "-wle", "print bpi(2000)"]
      restartPolicy: OnFailure     (4)
  1. Optional value for how many pod replicas a job should run in parallel; defaults to completions.

  2. Optional value for how many successful pod completions are needed to mark a job completed; defaults to one.

  3. Template for the pod the controller creates.

  4. The restart policy of the pod. This does not apply to the job controller. See Known Limitations for details.

You can also create and launch a job from a single command using oc run. The following command creates and launches the same job as specified in the previous example:

$ oc run pi --image=perl --replicas=1  --restart=OnFailure \
    --command -- perl -Mbignum=bpi -wle 'print bpi(2000)'

Known Limitations

The job specification restart policy only applies to the pods, and not the job controller. However, the job controller is hard-coded to keep retrying jobs to completion.

As such, restartPolicy: Never or --restart=Never results in the same behavior as restartPolicy: OnFailure or --restart=OnFailure. That is, when a job fails it is restarted automatically until it succeeds (or is manually discarded). The policy only sets which subsystem performs the restart.

With the Never policy, the job controller performs the restart. With each attempt, the job controller increments the number of failures in the job status and create new pods. This means that with each failed attempt, the number of pods increases.

With the OnFailure policy, kubelet performs the restart. Each attempt does not increment the number of failures in the job status. In addition, kubelet will retry failed jobs starting pods on the same nodes.

Scaling a Job

A job can be scaled up or down by using the oc scale command with the --replicas option, which, in the case of jobs, modifies the spec.parallelism parameter. This will result in modifying the number of pod replicas running in parallel, executing a job.

The following command uses the example job above, and scales it to three:

$ oc scale job pi --replicas=3
Scaling replication controllers also uses the oc scale command with the --replicas option, but instead changes the replicas parameter of a replication controller configuration.

Setting Maximum Duration

When defining a Job, you can define its maximum duration by setting the activeDeadlineSeconds field. It is specified in seconds and is not set by default. When not set, there is no maximum duration enforced.

The maximum duration is counted from the time when a first pod gets scheduled in the system, and defines how long a job can be active. It tracks overall time of an execution and is irrelevant to the number of completions (number of pod replicas needed to execute a task). After reaching the specified timeout, the job is terminated by OpenShift Origin.

The following example shows the part of a Job specifying activeDeadlineSeconds field for 30 minutes:

    activeDeadlineSeconds: 1800