- On compatible systems, a mounter's Unmount implementation is changed to not return an error when the specified target can be detected as not a mount point. On Linux, the behavior of detecting a mount point depends on `umount` command is validated when the mounter is created. Additionally, mount point checks will be skipped in CleanupMountPoint/CleanupMountWithForce if the mounter's Unmount having the changed behavior of not returning error when target is not...

- Scheduler: optionally check for available storage capacity before scheduling pods which have unbound volumes (alpha feature with the new `CSIStorageCapacity` feature gate, only works for CSI drivers and depends on support for the feature in a CSI driver deployment) ([#92387](https://github.com/kubernetes/kubernetes/pull/92387), [@pohly](https://github.com/pohly)) [SIG API Machinery, Apps, Auth, Scheduling, Storage and Testing]

    * kubeadm - Update generated etcd peer certificate to include localhost addresses for the default configuration.
    * kubeadm - Increase the manifest update timeout to make upgrades a bit more reliable.
* Kubernetes depends on v0.30.1 of cAdvisor ([#64987](https://github.com/kubernetes/kubernetes/pull/64987), [@dashpole](https://github.com/dashpole))

 selector mutation in
 a consistent way. This restriction may be lifted in the future, but
 it is likely that that selectors will remain immutable after the move to v1.
 You can continue to use code that depends on mutable selectors by calling
 the apps/v1beta1 API in this release, but you should start planning for code
 that does not depend on mutable selectors.

- On compatible systems, a mounter's Unmount implementation is changed to not return an error when the specified target can be detected as not a mount point. On Linux, the behavior of detecting a mount point depends on `umount` command is validated when the mounter is created. Additionally, mount point checks will be skipped in CleanupMountPoint/CleanupMountWithForce if the mounter's Unmount having the changed behavior of not returning error when target is not...

*   Improved DirectSession::Run() overhead and error checking. Feeding a value
    of the wrong type will now synchronously raise an INVALID_ARGUMENT error
    instead of asynchronously raising an INTERNAL error. Code that depends on
    the (undefined) behavior when feeding a tensor of the wrong type may need to
    be updated.
*   Added unreduced NONE, and reduced MEAN options for losses. Removed