Shrinking a PersistentVolumeClaim in Kubernetes

To shrink a PersistentVolumeClaim (PVC) in OpenShift or Kubernetes, we need to:

1. pause our deployments
2. copy its data to a smaller, temporary PVC
3. delete the original PVC
4. create a new PVC with the original name
5. copy the data from the temporary PVC to the new PVC
6. restart our deployments

First, we pause our deployments by scaling them down to zero replicas:

oc get deploy -o name -l app.kubernetes.io/instance=cube-fetalmri-org | xargs oc scale --replicas=0

Next, we need to create a temporary volume.

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: fetalmri-20240208-copy
spec:
  accessModes:
    - ReadWriteOnce
  volumeMode: Filesystem
  resources:
    requests:
      storage: 2Gi

To copy data between volumes, we need a container. The command here is sleep 99999999 so that we can use an interactive shell to do the copying instead.

apiVersion: v1
kind: Pod
metadata:
  name: copy
spec:
  containers:
    - name: sleep
      image: 'docker.io/instrumentisto/rsync-ssh:latest'
      command:
      - sleep
      - "99999999"
      volumeMounts:
      - mountPath: /orig
        name: orig
      - mountPath: /copy
        name: copy
  volumes:
  - name: orig
    persistentVolumeClaim:
      claimName: cube-fetalmri-org-storebase
  - name: copy
    persistentVolumeClaim:
      claimName: fetalmri-20240208-copy

Do the copying in an interactive shell:

oc exec -it copy -- rsync -ar --info=progress2 /orig/ /copy/

And verify:

oc exec -it copy -- ls /orig /copy
oc exec -it copy -- du -hs /orig /copy

Delete the pod:

oc delete --grace-period=1 pod copy

Now we can delete the original PVC and recreate a smaller one with the same name. The last steps are to copy the data back then restart our deployments. We can reuse the copy pod definition to copy the data back.