Reseach if we can eliminate the local-path-provisioner limitation that PV cannot be attached after rolling update

[ialidzhikov]

How to categorize this issue?

/area testing
/kind enhancement

What would you like to be added:
See

gardener-extension-registry-cache/test/e2e/cache/create_enabled_hibernate_reconcile_delete.go

Lines 60 to 63 in a4fdb06

	// PersistentVolumeClaims in local setup are provisioned by local-path-provisioner. local-path-provisioner creates a hostPath based persistent volume
	// on the Node. The provisioned PV is tightly bound to the Node. When the Node is deleted, the PV's hostPath directory is also deleted.
	// local-path-provisioner does not support moving PVC from one Node to another one (see https://github.com/rancher/local-path-provisioner/issues/31#issuecomment-690772828).
	// A local-path-provisioner PVC cannot deal with Node deletion/rollout.

.

Right now, we have TM tests for scenarios that include Node rolling update because we cannot test the scenario in e2e test because of the above limitation. It would be great if we can eliminate this limitation so that we replace the TM tests with the e2e alternatives.

Some ideas:

• Use in-place update for as strategy for rolling the Shoot Nodes. Not yet available, still in development at the time of writing this issue.
  • Help to tackle the issue for the CA rotation test but not for the hibernation test
• Use S3-compatible storage like min.io which is deployed in the Shoot control plane.
• Other ideas?

Why is this needed:
Replace TM tests with e2e ones.