fix(CSI-276): allow unpublish even if publish failed with stale file handle

[sergeyberezansky]

TL;DR

Improved error handling and logging in the NodeServer implementation.

What changed?

• Added handling for stale NFS mounts in NodeUnpublishVolume.

How to test?

1. Provision a snapshot-backed PVC and attach it to node on a 2.5.0-beta version of CSI plugin. You might hit a "stale file handle" error that causes the pod to not be able to access the volume contents. In such case, deletion of the pod will fail either.
2. Upgrade the plugin to latest version
3. Unpublish the volume by terminating the pod. Ensure that the PVC is deleted

Why make this change?

These changes ensure that pods with mis-attached PVCs (on version 2.5.0-beta2, 2.5.0-beta) may be detached and pods can be terminated. Without the fix, a node reboot might be required to remove the malfunctioning pods.

[graphite-app]

Graphite Automations

"Request reviewers once CI passes" took an action on this PR • (10/08/24)

1 reviewer was added to this PR based on Sergey Berezansky's automation.

[sergeyberezansky]

• chore(deps): update Go dependencies to latest #357
• fix(CSI-274): add sleep before mount if nfs was reconfigured #353
• fix(CSI-276): allow unpublish even if publish failed with stale file handle #356 👈
• fix(CSI-275): version of NFS is only set to V4 during NFS permission creation #354
• fix(CSI-273): remove rdirplus from mountoptions #355
• dev

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @sergeyberezansky and the rest of your teammates on Graphite

[sergeyberezansky]

Merge activity

• Oct 8, 11:10 AM EDT: @sergeyberezansky started a stack merge that includes this pull request via Graphite.
• Oct 8, 12:02 PM EDT: Graphite rebased this pull request as part of a merge.
• Oct 8, 12:04 PM EDT: Graphite rebased this pull request as part of a merge.
• Oct 8, 12:24 PM EDT: @sergeyberezansky merged this pull request with Graphite.