*   Upgrading the kubernetes-master units now results in staged upgrades just like the kubernetes-worker nodes. Use the upgrade action in order to continue the upgrade process on each unit such as juju run-action kubernetes-master/0 upgrade ([#55990](https://github.com/kubernetes/kubernetes/pull/55990),[ @hyperbolic2346](https://github.com/hyperbolic2346))

  - Added user-owned fields in `ResourceClaim` to specify resource requirements against each device capacity.
  - Added scheduler-owned field in `ResourceClaim.Status` to specify how much device capacity is reserved for a specific request.

  termination message log file. The termination message file is now properly
  readable for end users and has a maximum size (4k bytes) to prevent abuse.
  Each pod may have up to 12k bytes of termination messages before the contents
  of each will be truncated.
* Do not delete pod objects until all its compute resource footprint has been
  reclaimed.

- Kubelet: close a loophole where static pods could reference arbitrary ResourceClaims. The pods created by the kubelet then don't run due to a sanity check, but such references shouldn't be allowed regardless. ([#131875](https://github.com/kubernetes/kubernetes/pull/131875), [@pohly](https://github.com/pohly)) [SIG Apps, Auth and Node]

    * `kubectl create secret` has a new option `--from-env-file` that populates a secret from file which follows a key=val format for each line.

  * Printing/describe

  The API Server will no longer proxy non-101 responses for upgrade requests. This could break proxied backends (such as an extension API server) that respond to upgrade requests with a non-101 response code. ([#92941](https://github.com/kubernetes/kubernetes/pull/92941), [@tallclair](https://github.com/tallclair)) [SIG API Machinery]
 

    - `apiserver_flowcontrol_nominal_limit_seats`: Nominal number of execution seats configured for each priority level
    - `apiserver_flowcontrol_lower_limit_seats`: Configured lower bound on number of execution seats available to each priority level
    - `apiserver_flowcontrol_upper_limit_seats`: Configured upper bound on number of execution seats available to each priority level

  When the scheduling queue receives a cluster event, before moving each Pod from unschedulable pod pool to activeQ/backoffQ, it will call QueueingHintFn of plugins that rejected each Pod in the previous scheduling cycle.

### Bug or Regression

want to always pull the latest images before cluster init (as `Always` would do), run `kubeadm config images pull` before each `kubeadm init`. If you don't want the kubelet to pull any images at `kubeadm init` time, for example if you don't have an internet connection, you can also run `kubeadm config images pull` before `kubeadm init` or side-load the images some other way (such as `docker load -i image.tar`). Having the images locally cached will result in no pull at runtime, which makes it possible...