Deployment and Testing

Before we can test out our conversion, we’ll need to enable them conversion in our CRD:

Kubebuilder generates Kubernetes manifests under the config directory with webhook bits disabled. To enable them, we need to:

  • Enable patches/webhook_in_<kind>.yaml and patches/cainjection_in_<kind>.yaml in config/crd/kustomization.yaml file.

  • Enable ../certmanager and ../webhook directories under the bases section in config/default/kustomization.yaml file.

  • Enable manager_webhook_patch.yaml under the patches section in config/default/kustomization.yaml file.

  • Enable all the vars under the CERTMANAGER section in config/default/kustomization.yaml file.

Additionally, we’ll need to set the CRD_OPTIONS variable to just "crd", removing the trivialVersions option (this ensures that we actually generate validation for each version, instead of telling Kubernetes that they’re the same):

CRD_OPTIONS ?= "crd"

Now we have all our code changes and manifests in place, so let’s deploy it to the cluster and test it out.

You’ll need cert-manager installed (version 0.9.0+) unless you’ve got some other certificate management solution. The Kubebuilder team has tested the instructions in this tutorial with 0.9.0-alpha.0 release.

Once all our ducks are in a row with certificates, we can run make install deploy (as normal) to deploy all the bits (CRD, controller-manager deployment) onto the cluster.

Testing

Once all of the bits are up an running on the cluster with conversion enabled, we can test out our conversion by requesting different versions.

We’ll make a v2 version based on our v1 version (put it under config/samples)

apiVersion: batch.tutorial.kubebuilder.io/v2
kind: CronJob
metadata:
  name: cronjob-sample
spec:
  schedule:
    minute: "*/1"
  startingDeadlineSeconds: 60
  concurrencyPolicy: Allow # explicitly specify, but Allow is also default.
  jobTemplate:
    spec:
      template:
        spec:
          containers:
          - name: hello
            image: busybox
            args:
            - /bin/sh
            - -c
            - date; echo Hello from the Kubernetes cluster
          restartPolicy: OnFailure

Then, we can create it on the cluster:

kubectl apply -f config/samples/batch_v2_cronjob.yaml

If we’ve done everything correctly, it should create successfully, and we should be able to fetch it using both the v2 resource

kubectl get cronjobs.v2.batch.tutorial.kubebuilder.io -o yaml
apiVersion: batch.tutorial.kubebuilder.io/v2
kind: CronJob
metadata:
  name: cronjob-sample
spec:
  schedule:
    minute: "*/1"
  startingDeadlineSeconds: 60
  concurrencyPolicy: Allow # explicitly specify, but Allow is also default.
  jobTemplate:
    spec:
      template:
        spec:
          containers:
          - name: hello
            image: busybox
            args:
            - /bin/sh
            - -c
            - date; echo Hello from the Kubernetes cluster
          restartPolicy: OnFailure

and the v1 resource

kubectl get cronjobs.v1.batch.tutorial.kubebuilder.io -o yaml
apiVersion: batch.tutorial.kubebuilder.io/v2
kind: CronJob
metadata:
  name: cronjob-sample
spec:
  schedule:
    minute: "*/1"
  startingDeadlineSeconds: 60
  concurrencyPolicy: Allow # explicitly specify, but Allow is also default.
  jobTemplate:
    spec:
      template:
        spec:
          containers:
          - name: hello
            image: busybox
            args:
            - /bin/sh
            - -c
            - date; echo Hello from the Kubernetes cluster
          restartPolicy: OnFailure

Both should be filled out, and look equivalent to our v2 and v1 samples, respectively. Notice that each has a different API version.

Finally, if we wait a bit, we should notice that our CronJob continues to reconcile, even though our controller is written against our v1 API version.

Troubleshooting

steps for troubleshooting