[ModifyVolume] No VolumeModifyFailed event emitted when target VAC does not exist

[AndrewSirenko]

No VolumeModifyFailed nor VolumeModify event emitted when PVC is updated to point at non-existent VAC.

I deem this as an external-resizer bug because there is no easy way for customer to see why their volume is not modifying other than checking status.modifyVolumeStatus.status and seeing Pending.

Error message in resizer container:

csi-resizer E0729 00:03:26.778429 1 modify_volume.go:91] Get VAC with vac name <VAC-NAME> in VACInformer cache failed: volumeattributesclass.storage.k8s.io "io6-class" not found

---

Reproduction steps:

1. Launch pod and associated PVC (with or without VolumeAttributeClassName pointing to existent VAC).
2. Patch PVC to non-existent VAC: kubectl patch pvc ebs-claim --patch '{"spec": {"volumeAttributesClassName": "io8-class"}}'
3. Describe PVC

[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

[AndrewSirenko]

/remove-lifecycle stale

[sunnylovestiramisu]

The Status should be set to Pending is the VAC does not exist. The discussion was that we do not want to reinforce the order of creating different objects(no requirement of VAC must be created before PVC etc.).

[AndrewSirenko]

Hi @sunnylovestiramisu, thanks for the reply!

The Status should be set to Pending is the VAC does not exist.

I understand that status should be set to Pending in this case.

I raised this issue about emitting an event on the PVC mentioning missing VAC, so cluster operator can more easily see why modification never happened.

It seems that we emit an event in the provisioning path upon missing VAC:

Warning  ProvisioningFailed    1s (x3 over 2s)  ebs.csi.aws.com_ebs-csi-controller-5dc4fd4ffb-m4sr2_faa91941-2a1c-4cef-828f-64ab5a9f09 failed to provision volume with StorageClass "ebs-gp3-default": volumeattributesclasses.storage.k8s.io "gp4" not found

But a similar event is not emitted in modification path.