* Default value for expendable pod priority cutoff in GCP deployment of Cluster Autoscaler changed from 0 to -10.
    * action required: users deploying workloads with priority lower than 0 may want to use priority lower than -10 to avoid triggering scale-up.

### Other notable changes

* the template
* the template engine

* the type annotation
* the type hint

* the server worker
* the Uvicorn worker
* the Gunicorn Worker
* the worker process
* the worker class
* the workload

* the deployment
* to deploy

* the SDK
* the software development kit

* the `APIRouter`
* the `requirements.txt`
* the Bearer Token
* the breaking change
* the bug
* the button

### Bug or Regression

- Add `/sys/devices/virtual/powercap` to default masked paths. It avoids the potential security risk that the ability to read these files may offer a power-based sidechannel attack against any workloads running on the same kernel. ([#125970](https://github.com/kubernetes/kubernetes/pull/125970), [@carlory](https://github.com/carlory)) [SIG Node]
- Fix a bug that Pods could stuck in the unschedulable pod pool 

### Bug or Regression

- Add `/sys/devices/virtual/powercap` to default masked paths. It avoids the potential security risk that the ability to read these files may offer a power-based sidechannel attack against any workloads running on the same kernel. ([#125970](https://github.com/kubernetes/kubernetes/pull/125970), [@carlory](https://github.com/carlory)) [SIG Node]
- Fix a bug that Pods could stuck in the unschedulable pod pool 

- Introduced the `scheduling.k8s.io/v1alpha1` Workload API to express workload-level scheduling requirements and allow the kube-scheduler to act on them. ([#134564](https://github.com/kubernetes/kubernetes/pull/134564), [@macsko](https://github.com/macsko)) [SIG API Machinery, Apps, CLI, Etcd, Scheduling and Testing]...

## Major Features and Improvements

*   Support for Python3.9 has been added.
*   `tf.data`:
    *   `tf.data` service now supports strict round-robin reads, which is useful
        for synchronous training workloads where example sizes vary. With strict
        round robin reads, users can guarantee that consumers get similar-sized
        examples in the same step.

* Resolves spurious rollouts of workload controllers when upgrading the API server, due to incorrect defaulting of an alpha procMount field in pods ([#78883](https://github.com/kubernetes/kubernetes/pull/78883), [@liggitt](https://github.com/liggitt))

	// pick 10 nodes that precisely can send those 100 requests the first node
	// in the 10 node shard would coordinate between other 9 shards to get the
	// rest of the `99*9` requests.
	//
	// This essentially splits the workload properly and also allows for network
	// utilization to be optimal, instead of blindly throttling the way we are
	// doing below. However the changes that are needed here are a bit involved,

			if strings.HasSuffix(prefix, k) {
				return true
			}
		}
		return false
	}

	if hadoop && delimiter == SlashSeparator && maxKeys == 2 && marker == "" {
		// Optimization for Spark/Hadoop workload where spark sends a garbage
		// request of this kind
		//
		// GET /testbucket/?list-type=2&delimiter=%2F&max-keys=2&prefix=parquet%2F_SUCCESS%2F&fetch-owner=false
		//

- Removed extra pod creation expectations when daemonset fails to create pods in batches. ([#74856](https://github.com/kubernetes/kubernetes/pull/74856), [@draveness](https://github.com/draveness))
- Resolved spurious rollouts of workload controllers when upgrading the API server, due to incorrect defaulting of an alpha procMount field in pods. ([#78885](https://github.com/kubernetes/kubernetes/pull/78885), [@liggitt](https://github.com/liggitt))

### Auth