How can you scale Wordpress in Kubernetes cluster - using multiple pod replicas, - accessing a single PVC (persistent file storage)

Question

I'm learning Kubernetes, and trying to setup a cluster that could handle a single Wordpress site with high traffic. From reading multiple examples online from both Google Cloud and Kubernetes.io - they all set the "accessMode" - "readWriteOnce" when creating the PVCs.

Does this mean if I scaled the Wordpress Deployment to use multiple replicas, they all use the same single PVC to store persistent data - read/write data. (Just like they use the single DB instance?)

The google example here only uses a single-replica, single-db instance - https://cloud.google.com/kubernetes-engine/docs/tutorials/persistent-disk

My question is how do you handle persistent storage on a multiple-replica instance?

Answer 1

ReadWriteOnce means all replicas will use the same volume and therefore they will all run on one node. This can be suboptimal.

You can set up ReadWriteMany volume (NFS, GlusterFS, CephFS and others) storage class that will allow multiple nodes to mount one volume.

Alternatively you can run your application as StatefulSet with volumeClaimTemplate which ensures that each replica will mount its own ReadWriteOnce volume.

Answer 2

If on AWS (and therefore blocked by the limitation of EBS volumes only mountable on a single instance at a time), another option here is setting up the Pod Affinity to schedule on the same node. Not ideal from an HA standpoint, but it is an option.

If after setting that up you start running into any wonky issues (eg unable to login to the admin, redirect loops, disappearing media), I wrote a guide on some of the more common issues people run into when running Wordpress on Kubernetes, might be worth having a look!