This question follows on from this one regarding PV, PVC's and shared volumes.
I'm trying to get my head around what the best practices are regarding the provisioning of PV's for a NFS shared volume in a K8/OpenShift cluster.
As an example, let's say I have a 100GB NFS volume shared at /storage. What is the recommended way to carve that volume up as PV's?
The fact that you cannot target PVC's to specific PV's (well, until PersistentVolumeSelector is available) freaks me out a bit, in that my natural inclination is to want to target PVC's to specific PV's based on application need.
Would love to hear how the community is handling this in production environments?
I've seen people use a mix of strategies, but the former (carving into tiers of sizes) is a good place to start. You can use an annotation to target a PV to a PVC if you are planning app and PV at the same time, but if you aren't using dynamic provisioning the tiered sizes will at least give you a chance to split those up.
Mostly this comes down to planning what users need - NFS at least benefits that you can lie about size up front ant manage quota on the backend. I know of a few people who have done very very small NFS volumes (100M) and simply scaled those up as necessary.