Kubernetes 101, part VII, jobs and cronjobs

Our previous article, we learned how DaemonSets can effectively collect data from Kubernetes nodes, allowing data to be structured and sent to appropriate tooling.

In this post, we'll delve into the topic of running a single job in Kubernetes through the use of Kubernetes Jobs.

Furthermore, we'll learn how Kubernetes enables the scheduling of jobs to be executed regularly, through the use of Cronjobs.

---

Job

Kubernetes Job objects incorporate a Job controller that creates a Pod from the spec provided, enabling it to execute an arbitrary command.

That said, the YAML file looks like the following:

kind: Job
apiVersion: batch/v1
metadata:
  name: sleeper
spec:
  template:
    spec:
      restartPolicy: Never
      containers:
      - name: sleeper
        image: debian
        command: ["sleep", "15"]

We can check that the job sleeper was created:

$ kubectl get jobs

NAME      COMPLETIONS   DURATION   AGE
sleeper   0/1           5s        5s

And a Pod was started from the job created above:

$ kubectl get pods

NAME                    READY   STATUS    RESTARTS   AGE
sleeper-8mmtg           1/1     Running   0          8s

After the Pod is finished successfully, it goes to the Completed status:

$ kubectl get pods

NAME                    READY   STATUS    RESTARTS   AGE
sleeper-8mmtg           0/1     Completed   0          35s

And the Job completions is updated to 1/1:

$ kubectl get jobs

NAME      COMPLETIONS   DURATION   AGE
sleeper   1/1           35s        35s

Below is a visual representation of how Kubernetes handles Jobs, supported by a Pod that executes the command provided:

However, not rare we have to run a job regularly.

In UNIX-like systems, a program called crontab enables to run regularly, given a syntax for scheduling, an arbitrary command. Kubernetes allows Jobs to be scheduled in the same manner, using the the crontab syntax.

Meet Kubernetes CronJobs.

---

CronJob

In short, a Cronjob will regularly start a new Job, from which a new Pod will be started.

Here's an illustration of how CronJobs work in Kubernetes:

Below is the YAML representation of it:

kind: CronJob
apiVersion: batch/v1
metadata:
  name: sleeper
spec:
  schedule: "*/1 * * * *"
  jobTemplate:
    spec:
      template:
        spec:
          restartPolicy: Never
          containers:
          - name: sleeper
            image: debian
            command: ["sleep", "15"]

• in the schedule key we provide the crontab syntax
• the jobTemplate is exactly the Job specification that will run a Pod

Let's confirm that Kubernetes created a Cronjob object:

$ kubectl get cronjobs

NAME      SCHEDULE      SUSPEND   ACTIVE   LAST SCHEDULE   AGE
sleeper   */1 * * * *   False     0        52s             9m29s

From the CronJob, a Job was created:

$ kubectl get jobs

NAME               COMPLETIONS   DURATION   AGE
sleeper-28046804   0/1           22s        35s

And a Pod was started:

$ kubectl get pods

NAME                     READY   STATUS      RESTARTS   AGE
sleeper-28046805-hqkwb   1/1     Running     0          40s

After a couple of minutes, we see that the CronJob created 3 Jobs which, by the way, is the default number of Jobs that Kubernetes will keep in the cluster (we can change that number at our need successfulJobsHistoryLimit):

$ kubectl get jobs

NAME               COMPLETIONS   DURATION   AGE
sleeper-28046806   1/1           21s        3m10s
sleeper-28046807   1/1           21s        2m10s
sleeper-28046808   1/1           21s        70s

Therefore, each Job will start a new Pod:

$ kubectl get pods

NAME                     READY   STATUS      RESTARTS   AGE
sleeper-28046808-szdw7   0/1     Completed   0          2m26s
sleeper-28046809-fhxk8   0/1     Completed   0          86s
sleeper-28046810-2gvds   0/1     Completed   0          26s

---

Wrapping Up

This post demonstrated how to run arbitrary commands using the Kubernetes Job object.

Furthermore, we have learned that Kubernetes CronJob objects allow for the scheduling of Jobs to be executed regularly.

So far, we have covered the primary Kubernetes workload objects, namely ReplicaSet, Deployment, StatefulSet, DaemonSet, Job and CronJob.

In the following posts, we'll explore how these objects connect with each other within the cluster while also gaining an understanding of Kubernetes networking.

---

This post was written with the assistance of ChatGPT, which helped with some "eye candy" on grammar.

Comments

[Douglas Ritter]

This series is so good. I'been working for years with Kubernetes APIs wrapped in some in-house tooling before reading this and better understanding what I actually do in a daily basis. Thank you very much for writing this. I will recommend this reading to everyone I can.

[Leandro Proença]

I’m happy hearing that from you!