- Scores from PodTopologySpreading have reduced differentiation as maxSkew increases. ([#90820](https://github.com/kubernetes/kubernetes/pull/90820), [@alculquicondor](https://github.com/alculquicondor)) [SIG Scheduling]

  can also be scheduled to nodes with `GA` labels. ([#118923](https://github.com/kubernetes/kubernetes/pull/118923), [@AxeZhan](https://github.com/AxeZhan))
- Updated the generic apiserver library to produce an error if a new API server is configured with support for a data format other than JSON, YAML, or Protobuf. ([#121325](https://github.com/kubernetes/kubernetes/pull/121325), [@benluddy](https://github.com/benluddy)) [SIG API Machinery]

- Graduated the `LegacyServiceAccountTokenTracking` feature gate to Beta. The usage of auto-generated secret-based service account token now produces warnings by default, and relevant Secrets are labeled with a last-used timestamp (label key `kubernetes.io/legacy-token-last-used`). ([#114523](https://github.com/kubernetes/kubernetes/pull/114523), [@zshihang](https://github.com/zshihang)) [SIG...

- Promoted the Topology Manager policy option `max-allowable-numa-nodes` to GA version. ([#134614](https://github.com/kubernetes/kubernetes/pull/134614), [@ffromani](https://github.com/ffromani))
- Reduced event spam during volume operation errors in the Portworx in-tree driver. ([#135081](https://github.com/kubernetes/kubernetes/pull/135081), [@gohilankit](https://github.com/gohilankit))

- Kubeadm: fix the bug that configurable KubernetesVersion not respected during kubeadm join ([#111022](https://github.com/kubernetes/kubernetes/pull/111022), [@SataQiu](https://github.com/SataQiu)) [SIG Cluster Lifecycle]
- Reduced time taken to sync proxy rules on Windows kube-proxy with kernelspace mode ([#110702](https://github.com/kubernetes/kubernetes/pull/110702), [@daschott](https://github.com/daschott)) [SIG Network and Windows]

- An inefficient lock in EndpointSlice controller metrics cache has been reworked. Network programming latency may be significantly reduced in certain scenarios, especially in clusters with a large number of Services. ([#107091](https://github.com/kubernetes/kubernetes/pull/107091), [@robscott](https://github.com/robscott))

* 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))
* Reduced excessive logging from fluentd-gcp-scaler. ([#68837](https://github.com/kubernetes/kubernetes/pull/68837), [@x13n](https://github.com/x13n))

* Clarify function name in example in docs. PR [#1121](https://github.com/tiangolo/fastapi/pull/1121) by [@tmsick](https://github.com/tmsick).
* Add external link [Apache Kafka producer and consumer with FastAPI and aiokafka](https://iwpnd.pw/articles/2020-03/apache-kafka-fastapi-geostream) to docs. PR [#1112](https://github.com/tiangolo/fastapi/pull/1112) by [@iwpnd](https://github.com/iwpnd).

### Bug or Regression

- An inefficient lock in EndpointSlice controller metrics cache has been reworked. Network programming latency may be significantly reduced in certain scenarios, especially in clusters with a large number of Services. ([#107169](https://github.com/kubernetes/kubernetes/pull/107169), [@robscott](https://github.com/robscott)) [SIG Apps and Network]

 - Reduced state change noise when volume expansion fails. Also mark certain failures as infeasible.
  
  ACTION REQUIRED:  If you are using the `RecoverVolumeExpansionFailure` alpha feature gate