/priority important-soon
/assign @aojea @BenTheElder

Changelog suggestion (untested for accuracy; this is based on my guesses as to what the current text means)

Removed the last remaining in-tree code for integrating with GCP as cloud provider (cloud controller manager),
and code fetching GCP credentials used during container image pulls.
Please use the external cloud provider and credential provider from https://github.com/kubernetes/cloud-provider-gcp instead.

ℹ️ It's possible the removed code was for kubectl credential exec plugins instead, I haven't checked.

Hi! This change broke the autoscaler tests https://testgrid.k8s.io/sig-autoscaling-cluster-autoscaler#gci-gce-autoscaling - I started looking at them just recently (just before they broke as there were some constantly failing ones) and it seems that they might have missed some required migration. Since KEP does not mention any required migration for tests could you have a look and try pointing out what these tests are missing? They are probably missing some switch to external provider, but I am not sure where it should be made. Could you link for example migration of other tests so that I can apply the same logic to these tests?

This PR switches the provider implement to NULLProvider, it means that the provider does not do anything useful, it's there only to provide valid --provider cmdline option to allow testing of CSI migration tests of kubernetes.io/pd volume plugin.

FYI:

• kubernetes/test/e2e/framework/providers/gce/gce.go

  Lines 27 to 29 in a07d3c4

  	func newProvider() (framework.ProviderInterface, error) {
  	return &framework.NullProvider{}, nil
  	}

• (aws)

  kubernetes/test/e2e/framework/providers/aws/aws.go

  Lines 31 to 34 in a07d3c4

  	// Provider is a structure to handle AWS clouds for e2e testing
  	// It does not do anything useful, it's there only to provide valid
  	// --provider=aws cmdline option to allow testing of CSI migration
  	// tests of kubernetes.io/aws-ebs volume plugin.

e2e tests:

Hi @jbtk, Do those tests must be run on a cluster with GCE, or can be run on a kind cluster? I saw that @pohly is raising a PR #123078. In that PR, it uses kind cluster to run some tests related to cluster autoscaler and DRA.

cc @dims

should we revert this PR or just skip e2e tests util it can use kind cluster?

• kubernetes/test/e2e/autoscaling/cluster_size_autoscaling.go

  Line 109 in a07d3c4

  e2eskipper.SkipUnlessProviderIs("gce", "gke")

Learnt from the release notes: we may need to install https://github.com/kubernetes/cloud-provider-gcp before running the ginkgo test on GCP.

Hi! This change broke the autoscaler tests https://testgrid.k8s.io/sig-autoscaling-cluster-autoscaler#gci-gce-autoscaling - I started looking at them just recently (just before they broke as there were some constantly failing ones) and it seems that they might have missed some required migration. Since KEP does not mention any required migration for tests could you have a look and try pointing out what these tests are missing? They are probably missing some switch to external provider, but I am not sure where it should be made. Could you link for example migration of other tests so that I can apply the same logic to these tests?

These tests that are cloud provider specific have to be moved to the corresponding cloud-provider repo @jbtk , please reach out to @seans3 and @BenTheElder

kubernetes/cloud-provider-gcp#683

Learnt from the release notes: we may need to install https://github.com/kubernetes/cloud-provider-gcp before running the ginkgo test on GCP.

@carlory these tests will not work from this repo since this change

/triage accepted