- LimitRange and Endpoints resources can be created via an update API call if the object does not already exist. When this occurs, an authorization check is now made to ensure the user making the API call is authorized to create the object. In previous releases, only an update authorization check was performed. ([#65150](https://github.com/kubernetes/kubernetes/pull/65150), [@jennybuc...

LIST calls into multiple smaller chunks. A client can specify a limit to the number of results to return. If more results exist, a token is returned that allows the client to continue the previous list call repeatedly until all results are retrieved.  The resulting list is identical to a list call that does not perform chunking, thanks to capabilities provided by etcd3.  This allows the server to use less memory and CPU when very large lists are returned. This feature is gated as APIListChunking...

- Fixed panic when accessing CustomResources of a CRD with `x-kubernetes-int-or-string`. ([#83787](https://github.com/kubernetes/kubernetes/pull/83787), [@sttts](https://github.com/sttts))

  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:

    * inside of which a write lock was intended to prevent races against
    * VM creation, but the usage of the sync.RWMutex type is incorrect,
    * plus is placed in a defer call so the locking is effectively
    * nonexistent.  This patch adds an anonymous function call inside
    * the loop to correctly handle acquiring the write lock with deferred
    * unlock to run at that function's completion, following the normal

- When using `kubectl replace` (or the equivalent API call) on a Service, the caller no longer needs to do a read-modify-write cycle to fetch the allocated values for `.spec.clusterIP` and `.spec.ports[].nodePort`.  Instead the API server will automatically carry these forward from the original object when...

calls into multiple smaller chunks.  A client can specify a limit to the number of results to return, and if more results exist a token will be returned that allows the client to continue the previous list call repeatedly until all results are retrieved.  The resulting list is identical to a list call that does not perform chunking thanks to capabilities provided by etcd3.  This allows the server to use less memory and CPU responding with very large lists.  This feature is gated as APIListChunking and...

- Added apiserver_envelope_encryption_invalid_key_id_from_status_total to measure number of times an invalid keyID is returned by the Status RPC call. ([#115846](https://github.com/kubernetes/kubernetes/pull/115846), [@ritazh](https://github.com/ritazh)) [SIG API Machinery and Auth]

    * try to only update vm if detach a non-existing disk when got <200, error> after detach disk operation
* fix pod stuck issue due to corrupt mnt point in flexvol plugin, call Unmount if PathExists returns any error ([#75234](https://github.com/kubernetes/kubernetes/pull/75234), [@andyzhangx](https://github.com/andyzhangx))

    * inside of which a write lock was intended to prevent races against
    * VM creation, but the usage of the sync.RWMutex type is incorrect,
    * plus is placed in a defer call so the locking is effectively
    * nonexistent.  This patch adds an anonymous function call inside
    * the loop to correctly handle acquiring the write lock with deferred
    * unlock to run at that function's completion, following the normal