tensorflow::core::CppShapeInferenceResult::HandleData handle_data;
  if (!handle_data.ParseFromArray(proto, proto_len)) {
    status->status =
        absl::InvalidArgumentError("Couldn't deserialize HandleData proto");
    return;
  }
  DCHECK(handle_data.is_set());

  tensorflow::mutex_lock l(graph->mu);
  tensorflow::shape_inference::InferenceContext* ic =

- `kubeadm`: improved the logic that checks whether a `systemd` service exists. ([#120514](https://github.com/kubernetes/kubernetes/pull/120514), [@fengxsong](https://github.com/fengxsong))
- `kubeadm`: will now use universal deserializer to decode static pod. ([#120549](https://github.com/kubernetes/kubernetes/pull/120549), [@pacoxu](https://github.com/pacoxu))
- `kubectl prune v2`: Switched annotation from `contains-group-resources` to `contains-group-kinds`,

	wg.Add(2)
	go publish(ReportMetrics(GlobalContext, c.metricsGroups))
	go publish(globalNotificationSys.GetClusterMetrics(GlobalContext))
	wg.Wait()
}

// ReportMetrics reports serialized metrics to the channel passed for the metrics generated.
func ReportMetrics(ctx context.Context, metricsGroups []*MetricsGroupV2) <-chan MetricV2 {
	ch := make(chan MetricV2)
	go func() {
		defer xioutil.SafeClose(ch)

- The cloud config for Azure cloud provider can now be initialized from Kubernetes secret azure-cloud-provider in kube-system namespace
    - the secret is a serialized version of `azure.json` file with key cloud-config. And the secret name is azure-cloud-provider.

   * opportunistically, just to simplify some locking and avoid separate construction.
   */
  @SuppressWarnings("serial") // This class is never serialized.
  static final class Segment<K, V> extends ReentrantLock {

    /*
     * TODO(fry): Consider copying variables (like evictsBySize) from outer class into this class.

- Fixed a bug in the TopologyManager. Previously, the TopologyManager would only guarantee alignment if container creation was serialized in some way. Alignment is now guaranteed under all scenarios of container creation. ([#87759](https://github.com/kubernetes/kubernetes/pull/87759), [@klueska](https://github.com/klueska)) [SIG Node]

    `tf.name_scope` and `tf.variable_scope`. The new argument order of
    `tf.variable_scope` is incompatible with previous versions.
*   Introducing `core/util/tensor_bundle` module: a module to efficiently
    serialize/deserialize tensors to disk. Will be used in TF's new checkpoint
    format.
*   Added tf.svd for computing the singular value decomposition (SVD) of dense
    matrices or batches of matrices (CPU only).

* if kube-apiserver is started with `--storage-backend=etcd2`, the media type `application/json` is used. ([#43122](https://github.com/kubernetes/kubernetes/pull/43122), [@liggitt](https://github.com/liggitt))
* API fields that previously serialized null arrays as `null` and empty arrays as `[]` no longer distinguish between those values and always output `[]` when serializing to JSON. ([#43422](https://github.com/kubernetes/kubernetes/pull/43422), [@liggitt](https://github.com/liggitt))

- Fixed a data race that could occur when a single Go type was serialized to CBOR concurrently for the first time within a program. ([#129170](https://github.com/kubernetes/kubernetes/pull/129170), [@benluddy](https://github.com/benluddy)) [SIG API Machinery]

    * action required: Cloud providers other than GCE might want to update their deployments or sample yaml files to reuse the role created via add-on.

### Other notable changes

* Kubernetes json deserializer is now case-sensitive to restore compatibility with pre-1.8 servers. ([#65157](https://github.com/kubernetes/kubernetes/pull/65157), [@caesarxuchao](https://github.com/caesarxuchao))