* when updating a service or when node changes occur.
* Fix scheduling starvation of pods in cluster with large number of unschedulable pods. ([#72619](https://github.com/kubernetes/kubernetes/pull/72619), [@everpeace](https://github.com/everpeace))

* fix UnmountDevice failure on Windows ([#68608](https://github.com/kubernetes/kubernetes/pull/68608), [@andyzhangx](https://github.com/andyzhangx))
* adjusted http/2 buffer sizes for apiservers to prevent starvation issues between concurrent streams ([#67902](https://github.com/kubernetes/kubernetes/pull/67902), [@liggitt](https://github.com/liggitt))

* Improve efficiency of preemption logic in clusters with many pending pods. ([#72895](https://github.com/kubernetes/kubernetes/pull/72895), [@bsalamat](https://github.com/bsalamat))
* Fix scheduling starvation of pods in cluster with large number of unschedulable pods. ([#72619](https://github.com/kubernetes/kubernetes/pull/72619), [@everpeace](https://github.com/everpeace))

* kubectl: fixed an issue with "too old resource version" errors continuously appearing when calling `kubectl delete` ([#72825](https://github.com/kubernetes/kubernetes/pull/72825), [@liggitt](https://github.com/liggitt))
* Fix scheduling starvation of pods in cluster with large number of unschedulable pods. ([#72619](https://github.com/kubernetes/kubernetes/pull/72619), [@everpeace](https://github.com/everpeace))

* Fix scheduling starvation of pods in cluster with large number of unschedulable pods. ([#72619](https://github.com/kubernetes/kubernetes/pull/72619), [@everpeace](https://github.com/everpeace))

- Fixes an issue calculating total CPU usage reported for Windows nodes ([#122999](https://github.com/kubernetes/kubernetes/pull/122999), [@marosset](https://github.com/marosset)) [SIG Node and Windows]
- Prevent watch cache starvation by moving its watch to separate RPC and add a SeparateCacheWatchRPC feature flag to disable this behavior ([#123694](https://github.com/kubernetes/kubernetes/pull/123694), [@mengqiy](https://github.com/mengqiy)) [SIG API Machinery]

### Bug or Regression

- Fix component status calling etcd health endpoint over http which exposed kubernetes to the risk of complete watch starvation and is inconsistent with other etcd probing done by kube-apiserver. ([#119038](https://github.com/kubernetes/kubernetes/pull/119038), [@serathius](https://github.com/serathius)) [SIG API Machinery]

- Prevented watch cache starvation by moving its watch to separate RPC. Added a `SeparateCacheWatchRPC` feature flag to disable this behavior. ([#123532](https://github.com/kubernetes/kubernetes/pull/123532), [@serathius](https://github.com/serathius))

- When feature gate `JobTrackingWithFinalizers` is enabled:
  - Limit the number of Pods tracked in a single Job sync to avoid starvation of small Jobs.

* Get public IP for Azure vmss nodes. ([#68498](https://github.com/kubernetes/kubernetes/pull/68498), [@feiskyer](https://github.com/feiskyer))
* adjusted http/2 buffer sizes for apiservers to prevent starvation issues between concurrent streams ([#67902](https://github.com/kubernetes/kubernetes/pull/67902), [@liggitt](https://github.com/liggitt))