- `Scheduler` now registers event handlers dynamically. ([#101394](https://github.com/kubernetes/kubernetes/pull/101394), [@Huang-Wei](https://github.com/Huang-Wei))
- `kubectl`: Enable using protocol buffers to request Metrics API. ([#102039](https://github.com/kubernetes/kubernetes/pull/102039), [@serathius](https://github.com/serathius))

- The maximum length of the `CSINode` id field has increased to 256 bytes to match the CSI spec. ([#104160](https://github.com/kubernetes/kubernetes/pull/104160), [@pacoxu](https://github.com/pacoxu))
- Troubleshooting: informers log handlers that take more than 100 milliseconds to process an object if the `DeltaFIFO` queue starts to grow beyond 10 elements. ([#103917](https://github.com/kubernetes/kubernetes/pull/103917), [@aojea](https://github.com/aojea))

### Bug or Regression

- Fix missing delete events on informer re-lists to ensure all delete events are correctly emitted and using the latest known object state, so that all event handlers and stores always reflect the actual apiserver state as best as possible ([#115900](https://github.com/kubernetes/kubernetes/pull/115900), [@odinuge](https://github.com/odinuge)) [SIG API Machinery]

- Fix kube-apiserver startup to wait for APIServices to be installed into the HTTP handler before reporting readiness. ([#89147](https://github.com/kubernetes/kubernetes/pull/89147), [@sttts](https://github.com/sttts)) [SIG API Machinery]

- Updated an audit annotation key used by the `…/serviceaccounts/<name>/token` resource handler.

* Default resourceGroup should be used when the value of annotation azure-load-balancer-resource-group is an empty string. ([#79514](https://github.com/kubernetes/kubernetes/pull/79514), [@feiskyer](https://github.com/feiskyer))
* The CRD handler now properly re-creates stale CR storage to reflect CRD update. ([#79114](https://github.com/kubernetes/kubernetes/pull/79114), [@roycaihw](https://github.com/roycaihw))

    * Introduce a RuntimeClass v1beta1 API. This new beta API renames `runtimeHandler` to `handler`, makes it a required field, and cuts out the spec (handler is a top-level field).

  previous releases. When this happens, a `LifecycleHTTPFallback` event is recorded
  in the namespace of the pod and a `kubelet_lifecycle_handler_http_fallbacks_total`
  metric in the kubelet is incremented. Cluster administrators can opt out of the
  expanded lifecycle handler capabilities by setting
  `--feature-gates=ConsistentHTTPGetHandlers=false` in `kubelet`.'

- K8s.io/dynamic-resource-allocation/controller: UnsuitableNodes did not handle a mix of allocated and unallocated claims correctly. ([#120338](https://github.com/kubernetes/kubernetes/pull/120338), [@pohly](https://github.com/pohly)) [SIG Node]

* Detach and clear bad disk URI ([#58345](https://github.com/kubernetes/kubernetes/pull/58345), [@rootfs](https://github.com/rootfs))
* Correctly handle transient connection reset errors on GET requests from client library. ([#58520](https://github.com/kubernetes/kubernetes/pull/58520), [@porridge](https://github.com/porridge))