Leela Gangavarapu

- It's possible to implement it and would reduce a lot of recusive code for finding PVCs - However this was introduced to maximise the distribution afaik - If all...

- I understood the usecase, the issue is every pvc will end up under same dir hierarchy which may not be good for replica1 - afaik gluster shines when we...

@fastlorenzo from https://github.com/kadalu/kadalu/issues/747#issuecomment-1016244192 - when you say PVC name is deterministic do you mean combination of namespace and name of pvc that we supply in yaml or the generated name...

> Quota updates and PV metrics in the CSI provisioner may get affected by this mix of patterns. - And everything that expects PVC be under a random three paths...

> But in this case volname would be the PV name, not the PVC name, right? - yes, that's the issue, I don't see a way (yet!) supplying PVC name...

> If CSI can call K8s API - this opens up a lot of opportunities and I didn't put it forward as current implementation is k8s agnostic.

Thanks a lot for digging into this and this probably opens some optimisations. If you want to hang around, pls join slack for quick info exchange

@legendary-cookie afaik it's not possible with drivers implementing csi spec. Do you have the same usecase as the issue reporter or different one?

- I'll re-look this after #779 - IMHO i don't like to base size related ops from k8s perspective on simple quota yet - Sry if I'm being condesive here...

> The operator uses too much CPU after having a batch of nodes added and then removed. - are you using metrics exported by operator? - as you stated network...