* On GCP platforms, e2e testing now logs which OS images the cluster was found to have. ([#48310](https://github.com/kubernetes/kubernetes/pull/48310), [@abgworrall](https://github.com/abgworrall))

## Highlights

* Authorization:
  * **Alpha** RBAC authorization API group
* Federation
  * federation api group is now **beta**
  * Services from all federated clusters are now registered in Cloud DNS (AWS and GCP).
* Stateful Apps:
  * **alpha** PetSets manage stateful apps
  * **alpha** Init containers provide one-time setup for stateful containers
* Updating:
  * Retry Pod/RC updates in kubectl rolling-update.

  config_info_line('v1', 'Build with TensorFlow 1 API instead of TF 2 API.')

  print('Preconfigured Bazel build configs to DISABLE default on features:')
  config_info_line('nogcp', 'Disable GCP support.')
  config_info_line('nonccl', 'Disable NVIDIA NCCL support.')


if __name__ == '__main__':

- The in-tree GCE PD plugin `kubernetes.io/gce-pd` is now deprecated and will be removed in 1.21. Users that self-deploy Kubernetes on GCP should enable CSIMigration + CSIMigrationGCE features and install the GCE PD CSI Driver (https://github.com/kubernetes-sigs/gcp-compute-persistent-disk-csi-driver) to avoid disruption to existing Pod and PVC objects at that time. Users should start using the GCE PD CSI CSI Driver directly for any new volumes. (...

implement the analyzer is 11MB so we don't ship it with Elasticsearch by
default.

Another example is the `discovery-gce` plugin. It is *vital* to folks running
in [GCP](https://cloud.google.com/) but useless otherwise and it depends on a
dozen extra jars.

#### `qa`
Honestly this is kind of in flux and we're not 100% sure where we'll end up.
Right now the directory contains

## Dependencies

### Added
- buf.build/gen/go/bufbuild/protovalidate/protocolbuffers/go: 63bb56e
- github.com/GoogleCloudPlatform/opentelemetry-operations-go/detectors/gcp: [v1.26.0](https://github.com/GoogleCloudPlatform/opentelemetry-operations-go/tree/detectors/gcp/v1.26.0)
- github.com/bufbuild/protovalidate-go: [v0.9.1](https://github.com/bufbuild/protovalidate-go/tree/v0.9.1)

common:rbe_linux_cpu --copt=-Wno-gnu-offsetof-extensions
# Python config is the same across all containers because the binary is the same
common:rbe_linux_cpu --python_path="/usr/bin/python3"
# These you may need to change for your own GCP project.
common:rbe_linux_cpu --remote_instance_name=projects/tensorflow-testing/instances/default_instance

# Deprecated RBE config with non-hermetic toolchains.
common:rbe_linux_cpu_clang_local --config=rbe_linux_cpu

Related to this topic, the in-tree Azure and GCP authentication plugins have been [deprecated](https://github.com/kubernetes/kubernetes/pull/102181) in favor of out-of-tree implementations.

### Server-side Apply graduates to GA

## Changelog since v1.18.2

## Changes by Kind

### Bug or Regression
 - An issue preventing GCP cloud-controller-manager running out-of-cluster to initialize new Nodes is now fixed. ([#90057](https://github.com/kubernetes/kubernetes/pull/90057), [@ialidzhikov](https://github.com/ialidzhikov)) [SIG Apps and Cloud Provider]

- Removed the last remaining in-tree gcp cloud provider and credential provider.
  Please use the external cloud provider and credential provider from https://github.com/kubernetes/cloud-provider-gcp