consequence of this change the Kubelet now correctly takes into account the resources of running and terminating pods when deciding to accept new pods, since terminating pods are still holding on to those resources. This stricter handling may surface to end users as pod rejections when creating pods that are scheduled to mostly full nodes that have other terminating pods holding the resources the new pods need. The most likely error would be a pod set to `Failed` phase with reason set to `OutOfCpu` or `OutOfMemory`,...

- Fix a bug that Pods could stuck in the unschedulable pod pool 
  if they're rejected by PreEnqueue plugins that could change its result by a change in resources apart from Pods.
  
  DRA plugin is the only plugin that meets the criteria of the bug in in-tree, 
  and hence if you have `DynamicResourceAllocation` feature flag enabled, 

- Critical pods can now be created in namespaces other than kube-system. To limit critical pods to the kube-system namespace, cluster admins should create an admission configuration file limiting critical pods by default, and a matching quota object in the `kube-system` namespace permitting critical pods in that namespace. See https://kubernetes.io/docs/concepts/policy/resource-quotas/#limit-priority-clas...

  eg: `kubelet --node-ip 10.1.0.5,fd01::0005`. This is not yet supported for non-bare-metal
  clusters.
  
  In dual-stack clusters where nodes have dual-stack addresses, hostNetwork pods
  will now get dual-stack PodIPs. ([#95239](https://github.com/kubernetes/kubernetes/pull/95239), [@danwinship](https://github.com/danwinship)) [SIG Network and Node]

* Kubelet considers all static pods as critical. Static pods pass kubelet admission even if a node does not have enough resources. Users must ensure that they account for resources when creating static pods. ([#80491](https://github.com/kubernetes/kubernetes/pull/80491), [@hpandeycodeit](https://github.com/hpandeycodeit))

  Previously, when unschedulable Pods are updated, the scheduler always put Pods back to activeQ/backoffQ. But, actually not all updates to Pods make Pods schedulable, especially considering many scheduling constraints nowadays are immutable.

tolerations](https://kubernetes.io/docs/concepts/configuration/assign-pod-node/). This feature allows you to specify rules for "repelling" pods from nodes by default, which enables use cases like dedicated nodes and reserving nodes with special features for pods that need those features. We've also added a `NoExecute` taint type that evicts already-running pods, and an associated `tolerationSeconds` field to tolerations to delay the eviction for a specified amount of time. As before, taints are created...

* 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))

- Kube-scheduler now retries scheduling Pods rejected by the PodTopologySpread plugin when related Pods are created, deleted, updated, or when a node matches the specified topologyKey. ([#122195...

- There is now support for collecting pod logs under `/var/log/pods/NAMESPACE_NAME_UID` to stackdriver with `k8s_pod` resource type. ([#74502](https://github.com/kubernetes/kubernetes/pull/74502), [@Random-Liu](https://github.com/Random-Liu))