Hyperdisk smaller than 4GiB can not be created

[dfajmon]

When trying to create a hyperdisk smaller than 4GiB it fails.

Warning  ProvisioningFailed    20s (x10 over 5m1s)  pd.csi.storage.gke.io_ahafe-gcp-high-bhhd5-master-2.c.ocpstrat-1278.internal_ee5a8bd3-b1d9-4e9f-837d-73d624d90e91  failed to provision volume with StorageClass "balanced-storage": rpc error: code = InvalidArgument desc = CreateVolume failed: rpc error: code = InvalidArgument desc = CreateVolume failed to create single zonal disk pvc-4b0d2cc9-a677-4d36-9b34-bc14c8650826: failed to insert zonal disk: unknown Insert disk error: googleapi: Error 400: Disk size cannot be smaller than 4 GB for disk type hyperdisk-balanced., badRequest%!v(MISSING)

Resolution:
The driver should check if the size is valid, if the size is invalid it should round up to the closest.

[sunnylovestiramisu]

This will bypass quota check in Kubernetes, which is not the experience we want?

[dfajmon]

this has been discussed in upstream meeting and was accepted that it should be implemented
https://docs.google.com/document/d/1-oiNg5V_GtS_JBAEViVBhZ3BYVFlbSz70hreyaD7c5Y/edit?tab=t.0

cc @jsafrane

[sunnylovestiramisu]

Ignore this comment, I was mixing it with another issue

[msau42]

Rounding up results in unexpected costs to users, and also bypasses any k8s quota restrictions, so we are not going to round up to the minimum size.

For example, other storage systems like Filestore have a 1TB min size. It would definitely be a major unexpected cost to the customer if we rounded up the minimum size.

[jsafrane]

Users can cheat with quota today, they can request 1 byte of pd-ssd and the CSI driver provisions 1 GiB. That's why there is a quota on PVC count, not just PVC size. Why is 1 GiB OK and 4 GiB not? For example, the gcp-filestore CSI driver rounds to 1 TiB without hesitation and IMHO that's correct. Sorry, gcp filestore does not round to 1TB.

I understand the cost issues, however, no rounding imposes a problem to PVC authors. How can they write their operator / helm chart that works with all possible CSI drivers, esp. when their app needs just a small volume? This is not a hypothetical case, KubeVirt struggles with rounding rules of different clouds.

On the CSI side, a CSI driver gets a request with the minimum volume size, but with unbound maximum. A compliant driver should provision something.

[k8s-triage-robot]

The Kubernetes project currently lacks enough contributors to adequately respond to all issues.

This bot triages un-triaged issues according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Mark this issue as fresh with /remove-lifecycle stale
• Close this issue with /close
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle stale