they remain valid even after a new refreshed token is provided. The metric `serviceaccount_stale_tokens_total` can be used to monitor for workloads that are depending on the extended lifetime and are continuing to use tokens even after a refreshed token is provided to the container. If that metric indicates no existing workloads are depending on extended lifetimes, injected token lifetime can be shortened to 1 hour by starting `kube-apiserver` with `--service-account-extend-token-expiration=false`....

on a number of [Container Storage Interface (CSI) features](https://github.com/kubernetes/enhancements/issues/178) this quarter in anticipation of moving support for CSI from beta to GA in the next Kubernetes release. This includes graduating the dependent “mount namespace propagation” feature to GA, moving the Kubelet plugin registration mechanism to beta, adding alpha support for a new CSI driver registry as well as for topology, and adding a number of alpha features to support the use of CSI for...

* Fix a scalability issue where high rates of event writes degraded etcd performance. ([#64539](https://github.com/kubernetes/kubernetes/pull/64539), [@ccding](https://github.com/ccding))
* "kubectl delete" no longer waits for dependent objects to be deleted when removing parent resources ([#65908](https://github.com/kubernetes/kubernetes/pull/65908), [@juanvallejo](https://github.com/juanvallejo))

          <description>
            <![CDATA[
            Whether the directory described by this source element should be included in the build.
            This flag provides an easy way to include or exclude some sources depending, for example,
            o property values defined in profiles. A use case is including optional resources only
            when the user confirmed a license agreement.

### Garbage Collector

- DRA: Device taints enable DRA drivers or admins to mark device as unusable, which prevents allocating them. Pods may also get evicted at runtime if a device becomes unusable, depending on the severity of the taint and whether the claim tolerates the taint. ([#130447](https://github.com/kubernetes/kubernetes/pull/130447), [@pohly](https://github.com/pohly)) [SIG API Machinery, Apps, Architecture, Auth, Etcd, Instrumentation, Node,...

- DRA: scheduling pods is up to 16x faster, depending on the scenario. Scheduling throughput depends a lot on cluster utilization. It is higher for lightly loaded clusters with free resources and gets lower when the cluster utilization increases. ([#127277](https://github.com/kubernetes/kubernetes/pull/127277),...

- Introduce `node.kubernetes.io/exclude-balancer` and `node.kubernetes.io/exclude-disruption` labels in alpha to prevent cluster deployers from being dependent on the optional `node-role` labels which not all clusters may provide. ([#80238](https://github.com/kubernetes/kubernetes/pull/80238), [@smarterclayton](https://github.com/smarterclayton))

longer being set, and are being replaced by the more portable mechanism of the kubelet configuration file. The individual variables in the kube-env metadata entry were never meant to be a stable interface and this release note only applies if you are depending on them. ([#60020](https://github.com/kubernetes/kubernetes/pull/60020), [@roberthbailey](https://github.com/roberthbailey))

* kube-proxy: feature gates are now specified as a map when provided via a JSON or YAML KubeProxyConfiguration,...

  When the scheduling queue receives a cluster event, before moving each Pod from unschedulable pod pool to activeQ/backoffQ, it will call QueueingHintFn of plugins that rejected each Pod in the previous scheduling cycle.
  Depending on the value returned from QueueingHintFn, the scheduling queue changes how it queues each Pod:
  - if more than one QueueingHintFn returns QueueImmediately, it queues Pod to activeQ.