Fix VolumeAttachment garbage collection for migrated PVs

[timebertt]

What type of PR is this?

/kind bug

What this PR does / why we need it:

This PR builds on #101737 to also cleanup dangling VolumeAttachments after completed migration to CSI (like described in #102097).

Which issue(s) this PR fixes:

Fixes #102097

Special notes for your reviewer:

/sig storage

I validated the change on our clusters, where we observed the issue (AWS/GCP, k8s v1.18/v1.19). On startup, kube-controller-manager now marks the old VolumeAttachments as uncertain and later detaches them, e.g.:

{"log":"Marking volume attachment as uncertain as volume:\"kubernetes.io/csi/pd.csi.storage.gke.io^projects/UNSPECIFIED/zones/europe-west1-b/disks/pvc-some-id\" (\"some-node\") is not attached (Detached)","pid":"1","severity":"INFO","source":"attach_detach_controller.go:766"}
{"log":"attacherDetacher.DetachVolume started for volume \"pvc-some-id\" (UniqueName: \"kubernetes.io/csi/pd.csi.storage.gke.io^projects/UNSPECIFIED/zones/europe-west1-b/disks/pvc-some-id\") on node \"some-node\" This volume is not safe to detach, but maxWaitForUnmountDuration 6m0s expired, force detaching","pid":"1","severity":"WARN","source":"reconciler.go:222"}
{"log":"DetachVolume.Detach succeeded for volume \"pvc-some-id\" (UniqueName: \"kubernetes.io/csi/pd.csi.storage.gke.io^projects/UNSPECIFIED/zones/europe-west1-b/disks/pvc-some-id\") on node \"some-node\" ","pid":"1","severity":"INFO","source":"operation_generator.go:472"}

Does this PR introduce a user-facing change?

Fixed garbage collection of dangling VolumeAttachments for PersistentVolumes migrated to CSI on startup of kube-controller-manager.

Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:

[k8s-ci-robot]

Welcome @timebertt!

It looks like this is your first PR to kubernetes/kubernetes 🎉. Please refer to our pull request process documentation to help your PR have a smooth ride to approval.

You will be prompted by a bot to use commands during the review process. Do not be afraid to follow the prompts! It is okay to experiment. Here is the bot commands documentation.

You can also check if kubernetes/kubernetes has its own contribution guidelines.

You may want to refer to our testing guide if you run into trouble with your tests not passing.

If you are having difficulty getting your pull request seen, please follow the recommended escalation practices. Also, for tips and tricks in the contribution process you may want to read the Kubernetes contributor cheat sheet. We want to make sure your contribution gets all the attention it needs!

Thank you, and welcome to Kubernetes. 😃

[k8s-ci-robot]

Hi @timebertt. Thanks for your PR.

I'm waiting for a kubernetes member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work. Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[timebertt]

/cc @Jiawei0227
if you want to have a look :)

[Jiawei0227]

/ok-to-test

[Jiawei0227]

/lgtm
Thanks for the fix! I think this is looking good to me. Just want to understand the issue more, this only happens if there is indeed dangling attachment in the system, right? in a normal sequence, after CSI migration, the Pod should still able to detach this volume?

[Jiawei0227]

/retest

[Jiawei0227]

/retest

[timebertt]

Just want to understand the issue more, this only happens if there is indeed dangling attachment in the system, right? in a normal sequence, after CSI migration, the Pod should still able to detach this volume?

Yes, this bug is a real edge case, it doesn't happen in a normal sequence.

In all cases, where we observed the bug, kube-controller-manager was restarted while an attach/detach operation was in progress, e.g. because a StatefulSet pod was rolled/restarted (like described in #96617). What happens on startup then, is that it tries to process the old VolumeAttachment, but it can't find a responsible volume plugin (AWS/GCP is unregistered and CSI doesn't feel responsible as the volume was created by in-tree plugins), here:

kubernetes/pkg/controller/volume/attachdetach/attach_detach_controller.go

Lines 719 to 728 in be58307

	plugin, err := adc.volumePluginMgr.FindAttachablePluginBySpec(volumeSpec)
	if err != nil || plugin == nil {
	// Currently VA objects are created for CSI volumes only. nil plugin is unexpected, generate a warning
	klog.Warningf(
	"Skipping processing the volume %q on nodeName: %q, no attacher interface found. err=%v",
	*pvName,
	nodeName,
	err)
	continue
	}

If the pod was scheduled on another node, then the Volume won't be detached on the old node as the old VolumeAttachment is not cleaned up, which leads to exclusive attach errors.

This is quite hard to reproduce, because it has to happen in a very precise timing (which in large scale environments can of course happen more frequently).
Does this help? :)

[timebertt]

/retest

[Jiawei0227]

/retest

[Jiawei0227]

This is quite hard to reproduce, because it has to happen in a very precise timing (which in large scale environments can of course happen more frequently).
Does this help? :)

Thanks a lot! This is very helpful info.
/lgtm
/cc @gnufied
For approval.

[timebertt]

/retest

[timebertt]

/retest

[gnufied]

/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: gnufied, timebertt

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• pkg/controller/volume/attachdetach/OWNERS [gnufied]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[gnufied]

/triage accepted