### Network

    *   reflector_items_per_watch
    *   reflector_last_resource_version
    *   reflector_list_duration_seconds
    *   reflector_lists_total
    *   reflector_short_watches_total
    *   reflector_watch_duration_seconds
    *   reflector_watches_total
    * While this is a backwards-incompatible change, it would have been impossible to setup reliable monitoring around these metrics since the labels were not stable. 

* Added a feature gate for the plugin watcher. ([#64605](https://github.com/kubernetes/kubernetes/pull/64605), [@vikaschoudhary16](https://github.com/vikaschoudhary16))

- kube-apiserver: fixed a bug that could cause a goroutine leak if the apiserver encountered an encoding error serving a watch to a websocket watcher ([#84693](https://github.com/kubernetes/kubernetes/pull/84693), [@tedyu](https://github.com/tedyu))

    * Don't watch .mount cgroups to reduce number of inotify watches
    * Fix NVML initialization race condition
    * Fix brtfs disk metrics when using a subdirectory of a subvolume
* Bump version of prometheus-to-sd to 0.2.6 to decrease log noise and include latest security patches. ([#64964](https://github.com/kubernetes/kubernetes/pull/64964), [@loburm](https://github.com/loburm))

  
  - https://github.com/kubernetes-sigs/kustomize/pull/4911: Drop support for a very old, legacy style of patches. patches used to be allowed to be used as an alias for patchesStrategicMerge in kustomize v3. You now have to use patchesStrategicMerge explicitly, or update to the new syntax supported by patches. See examples in the PR description of https://github.com/kubernetes-sigs/kustomize/pull/4911.

- Fixes issue with websocket-based watches of Service objects not closing correctly on timeout ([#102541](https://github.com/kubernetes/kubernetes/pull/102541), [@liggitt](https://github.com/liggitt)) [SIG API Machinery and Testing]

## Dependencies

### Added

    *   reflector_items_per_watch
    *   reflector_last_resource_version
    *   reflector_list_duration_seconds
    *   reflector_lists_total
    *   reflector_short_watches_total
    *   reflector_watch_duration_seconds
    *   reflector_watches_total
    * While this is a backwards-incompatible change, it would have been impossible to setup reliable monitoring around these metrics since the labels were not stable. 

incorrect.  If watching objects with a filter, when an update was made that no longer matched the filter a DELETE event was correctly sent.  However, the object that was returned by that delete was not the (correct) version before the update, but instead, the newer version.  That meant the new object was not matched by the filter.  This was a regression from behavior between cached watches on the server side and uncached watches, and thus broke downstream API clients. ([#46223](https://github.com/kuberne...

- Ensure Job status updates are batched by 1s. This fixes an unlikely scenario when a sequence of immediately 
  completing pods could trigger a sequence of non-batched Job status updates. ([#118470](https://github.com/kubernetes/kubernetes/pull/118470), [@mimowo](https://github.com/mimowo)) [SIG Apps]