runOnce Daemonset Job has no releases yet. Stay tuned.

runOnce Daemonset Job for OpenShift (and Kubernetes)

Lets say you want to be able to run a job on a series of nodes with a specific label.

A Daemonset would let you run pods on different nodes, but not as a job
A Job will let you run something once, but there's no guarantee it'll run on all the nodes you want it to

This git repo can help...

A CronJob whose job is to create downstream Jobs - aka daemonset-jobs

In it's simplest form, 'runOnce Daemonset Job' is a Kubernetes CronJob that will kick off downstream Job(s), passing down things like environment variables, images, scripts etc, as needed.

A bit more detail: The main Job will create your yaml Job definitions, replacing variables as needed, and then creating the Job based on that definition.

How does it work?

First, we create a cronjob definition named daemonset-jobs.yaml. There's a sample in this repository at scripts/daemonset-job_sample.yaml. The file contains a typical Kubernetes definition for a cronjob and also includes any environment variables or other details you want to pass down to your daemonset-job.

You MUST update the fields that are commented out in the scripts/daemonset-job_sample.yaml file to meet your needs.

Check out this pretty image to better understand what's going on under the hood.

Variables

Before you get going, there's a few variables that MUST be specified in your main cronjob definition.

Required Variables

JOB_NAME : The name that the downstream job will use. The final job name will appended with a numerical value, depending on the number of nodes you run it on
JOB_TO_RUN : This is the yaml definition for the downstream Job you want to run. Example value for the key would be host-checks.yaml
RUN_SCRIPT : The script to run as CMD on the downstream Job's pod.
NODE_SELECTOR_KEY : The key to be used for the nodeSelector. A recommendation is to specify 'kubernetes.io/hostname' for this value as it would give the most control
NODE_SELECTOR_VALUE : A space-separated list of nodeSelector values to use. My recommendation is that this is list of node names. In combination with 'kubernetes.io/hostname' you can specify all the nodes you want the daemonset-job to run on by their hostname
IMAGE_TO_USE : This is the Docker image to use for the downstream Job.

Optional Variable(s)

DAEMONSETJOB_<ANYTHING> : Use this when you want the downstream Job to have environment variables. Any variable prepended with DAEMONSETJOB_ will be interpreted as an environment variable to be used by the downstream Job and the DAEMONSETJOB_ will be removed from the final variable name.

Here's an example of what your environment variables might look like in your main cronJob definition.

kind: CronJob
spec:
  jobTemplate:
    spec:
      template:
        spec:
          containers:
            - env:
              - name: JOB_NAME
                value: host-disk-checks
              - name: JOB_TO_RUN
                value: host-checks.yaml
              - name: RUN_SCRIPT
                value: host-checks.sh
              - name: NODE_SELECTOR_KEY
                value: kubernetes.io/hostname
              - name: NODE_SELECTOR_VALUE
                value: app01.mydomain.com app02.mydomain.com opp03.mydomain.com
                  buildnode01.mydomain.com infra01.mydomain.com infra02.mydomain.com
                  infra03.mydomain.com logs01.mydomain.com logs02.mydomain.com
              - name: IMAGE_TO_USE
                value: docker-registry.default.svc:5000/monitor/ocp-ops-base-image:latest
              - name: DAEMONSETJOB_SOME_NUMBER
                value: "70"
              - name: DAEMONSETJOB_SOME_BOOLEAN
                value: 'true'
              - name: DAEMONSETJOB_SOME_STRING
                value: here's some text I want to pass into the downstream Job

In addition to this, you'll need to create a serviceaccount in order to kick off the jobs for you since the container will be using the oc apply command.

Here's an example:

oc create sa dsj --namespace <mynamespace>
oc policy add-role-to-user admin --serviceaccount=dsj --namespace <mynamespace> (you might want to choose something other than admin if your jobs don't need it)