- Added `singleProcessOOMKill` flag to the kubelet configuration. Setting that to true enable single process OOM killing in cgroups v2. In this mode, if a single process is OOM killed within a container, the remaining processes will not be OOM killed. ([#126096](https://github.com/kubernetes/kubernetes/pull/126096), [@utam0k](https://github.com/utam0k)) [SIG API Machinery, Node, Testing and Windows]

- A new flag called `custom` has been introduced in `kubectl debug`, allowing users to customize pre-defined profiles. ([#120346](https://github.com/kubernetes/kubernetes/pull/120346), [@ardaguclu](https://github.com/ardaguclu))

  then kube-proxy will send the traffic to the Terminating pod rather than dropping it. This
  patches up a race condition between when a pod is killed and when the external load
  balancer notices that it has been killed. ([#97238](https://github.com/kubernetes/kubernetes/pull/97238), [@andrewsykim](https://github.com/andrewsykim))

- Kubeadm: fixed a bug where an image is not pulled if there is an error with the sandbox image from CRI. ([#129607](https://github.com/kubernetes/kubernetes/pull/129607), [@neolit123](https://github.com/neolit123)) [SIG Cluster Lifecycle]

- Kubeadm: fix the bug that 'kubeadm upgrade' hangs in single node cluster ([#88434](https://github.com/kubernetes/kubernetes/pull/88434), [@SataQiu](https://github.com/SataQiu)) [SIG Cluster Lifecycle]
- Kubeadm: make sure images are pre-pulled even if a tag did not change but their contents changed ([#85603](https://github.com/kubernetes/kubernetes/pull/85603), [@bart0sh](https://github.com/bart0sh)) [SIG Cluster Lifecycle]

the list of authentication information that leads to their successful pulls. This data is persisted across reboots of the host and restarts of the kubelet.
  
  The kubelet ensures any image requiring credential verification is always pulled if authentication information from an image pull is not yet present, thus enforcing authentication / re-authentication. This means an image pull might be attempted even in cases where a pod requests the `IfNotPresent` image pull policy, and might...

* Implement envelope service with gRPC, so that KMS providers can be pulled out from API server. ([#55684](https://github.com/kubernetes/kubernetes/pull/55684), [@wu-qiang](https://github.com/wu-qiang))

- For raw block CSI volumes on Kubernetes, kubelet was incorrectly calling CSI NodeStageVolume for every single "map" (i.e. raw block "mount") operation for a volume already attached to the node. This PR ensures it is only called once per volume per node. ([#112403](https://github.com/kubernetes/kubernetes/pull/112403), [@akankshakumari393](https://github.com/akankshakumari393)) [SIG Storage]

## API Changes

- [CRI] Added a new field called `runtime_handler` into `PodSandbox` and `PodSandboxStatus` to track the `RuntimeClass` information of a pod. ([#73833](https://github.com/kubernetes/kubernetes/pull/73833), [@haiyanmeng](https://github.com/haiyanmeng))

  2. if there's a transient error happened during "provisioning/deletion", i.e., a volume is still in-use while a deleteVolume has been called, original `storage_operation_duration_seconds` will NOT wait until a volume has been finally deleted before reporting an inaccurate latency metric. The newly implemented metric `volume_operation_total_seconds`, however, waits until a provisioning/deletion...