* kube-apiserver: requests to endpoints handled by unavailable extension API servers (as indicated by an `Available` condition of `false` in the registered APIService) now return `503` errors instead of `404` errors. ([#58070](https://github.com/kubernetes/kubernetes/pull/58070), [@weekface](https://github.com/weekface))

- The 404 request handler for the GCE Ingress load balancer now exports prometheus metrics, including:

  - `http_404_request_total` (the number of 404 requests handled)
  - `http_404_request_duration_ms` (the amount of time the server took to respond in ms)

- Sync node status during kubelet node shutdown.
  Adds an pod admission handler that rejects new pods when the node is in progress of shutting down. ([#98005](https://github.com/kubernetes/kubernetes/pull/98005), [@wzshiming](https://github.com/wzshiming)) [SIG Node]

### Other notable changes

- EndpointSlice controller now handles terminating pods correctly and is better at preventing race conditions. ([#89118](https://github.com/kubernetes/kubernetes/pull/89118), [@robscott](https://github.com/robscott)) [SIG Apps and Network]

* 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))

- kube-scheduler removed `AzureDiskLimits` ,`CinderLimits` `EBSLimits` and `GCEPDLimits` plugin. Given the corresponding CSI driver reports how many volumes a node can handle in NodeGetInfoResponse, the kubelet stores this limit in CSINode and the scheduler then knows the limit of the driver on the node. Removed plugins AzureDiskLimits, CinderLimits, EBSLimits and GCEPDLimits if you explicitly enabled them in the scheduler...

- Kubelet(dra): correctly handles multiple ResourceClaims even if one is already prepared ([#136463](https://github.com/kubernetes/kubernetes/pull/136463), [@rogowski-piotr](https://github.com/rogowski-piotr)) [SIG Node and Testing]

- 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]

- Changed how the API server handles updates to `.spec.defaultBackend` of Ingress objects.