how to safely remove digitalocean csi plugin

Hey!

When removing the DigitalOcean CSI plugin, you want to ensure the process is handled methodically to avoid unintended consequences.

Before you prcoeed I would strongly recommend making sure that you actually have a working backup. You can for example use the DigitalOcean Kubernetes backups with SnapShooter:

https://www.digitalocean.com/blog/doks-backup-snapshooter

Here are some steps, along with command examples, to remove the CSI plugin safely:

1. Drain Kubernetes nodes (Optional but recommended for maintenance): This is to ensure that no work is being done on the node and no new work will be scheduled on it.

   kubectl drain <node-name> --ignore-daemonsets --delete-emptydir-data
   

2. Scale down Deployments/StatefulSets: If you have deployments or stateful sets using the CSI driver, scale them down.

   kubectl scale deployment <deployment-name> --replicas=0
   kubectl scale statefulset <statefulset-name> --replicas=0
   

3. Delete PersistentVolumeClaims (PVCs): Remove PVCs that are provisioned by the DigitalOcean CSI driver.

   kubectl delete pvc <pvc-name>
   

   After deleting the PVCs, make sure that the associated PVs are also deleted. You can check using:

   kubectl get pv
   

   If the PVs still exist and you’re sure they’re not needed anymore (or you’ve backed up the data elsewhere), you can delete them:

   kubectl delete pv <pv-name>
   

4. Delete the CSI driver components: This includes the controller deployments, CSI driver daemonset, and any associated secrets or other resources.

   kubectl delete -f <csi-driver-deployment-file.yaml>
   kubectl delete ds -n kube-system <csi-driver-daemonset-name>
   kubectl delete secret -n kube-system <csi-driver-secret-name>
   

   These commands require you to know the exact names of the resources deployed by the CSI plugin. You can find them with:

   kubectl get all -n kube-system | grep csi
   

5. Delete CustomResourceDefinitions (CRDs): If the CSI plugin used CRDs, you’ll need to delete those as well.

   kubectl delete crd <crd-name>
   

6. Clean up ClusterRoleBindings and ClusterRoles: If the CSI plugin created any cluster roles or bindings, they should also be removed.

   kubectl delete clusterrolebinding <clusterrolebinding-name>
   kubectl delete clusterrole <clusterrole-name>
   

7. Review Storage Classes: Make sure that your NFS storage class is set as default if you no longer intend to use DigitalOcean’s block storage.

   kubectl patch storageclass <your-nfs-storageclass-name> -p '{"metadata": {"annotations":{"storageclass.kubernetes.io/is-default-class":"true"}}}'
   

8. Uncordon the nodes (If you drained them): Once the maintenance is complete, make sure to uncordon the nodes to allow scheduling workloads on them again.

   kubectl uncordon <node-name>
   

Best,

Bobby