local-cluster-pvc-change

Modify PVC Storage

As providers mature CSI drivers are becoming available. This scenario will migrate storage from a gp2 storageclass to a gp2-csi storageclass. The process is defined to ensure limited to no downtime when swiching the underlying storage.

Deploying the application

A Dokuwiki application will be deployed with a pvc using gp2.

oc create -f ./application/ -n dokuwiki

Deploying Scribe

We will use Scribe to manage the replication between our gp2 and gp2-csi storageclass.

To deploy Scribe install the Scribe Helm chart using the instructions (here)[https://scribe-replication.readthedocs.io/en/latest/installation/index.html#kubernetes-openshift]

Now that Scribe has been deployed we can define the replicationsource and replicationdestination objects.

oc create -f ./destination-rsync
oc create -f ./source-rsync

This will replicate the data from gp2 to a gp2-csi every 8 minutes. As it gets closer to the time to perform the migration. We will modify the replicationsource object to use a shorter increment of time. This is possibly because rsync replicates only the files that have changed allowing for replication to happen in the smaller replication window without worrying about replication jobs stacking up.

Modify the replicationsource to synchronize every 1 minute rather than 8.

vi ./source-rsync/replicationsource.yaml
oc replace -f ./source-rsync/replicationsource.yaml

Migration

When you believe that most if not all data has been replicated perform the following.

Pause the replicationdestination object by adding in the following line paused: true

vi ./destination-sync/replicationdestination.yaml
...redacted
spec:
  paused: true
  rsync:
    serviceType: ClusterIP
    copyMethod: None
...redacted

Replace the replicationdestination object.

oc replace -f ./destination-sync/replicationdestination.yaml

Modify the deployment to use the new pvc.

vi ./application/deployment.yaml
...redacted
      volumes:
      - name: dokuwiki-data
        persistentVolumeClaim:
          claimName: dokuwiki-csi

Apply the changes to the deployment which will cause a rollout of a new pod.

oc apply -f ../application/deployment.yaml

Due to how pods are deployed the application should stay online and gracefully transfer between the container using gp2 to the container using gp2-csi.