serving watch requests without a resourceVersion from the watch cache by default, as in <1.27 (disabling the change in #115096 by default). This mitigates the impact of an etcd watch bug (https://github.com/etcd-io/etcd/pull/17555). If the 1.27 change in #115096 to serve these requests from underlying storage is still desired despite the impact on watch stability, it can be re-enabled with a `WatchFromStorageWithoutResourceVersion` feature gate. ([#124006](https://github.com/kubernetes/kubernetes/pull/124006),...

* Respect fstype in Windows for azure disk ([#61267](https://github.com/kubernetes/kubernetes/pull/61267), [@andyzhangx](https://github.com/andyzhangx))
* Unready pods will no longer impact the number of desired replicas when using horizontal auto-scaling with external metrics or object metrics. ([#60886](https://github.com/kubernetes/kubernetes/pull/60886), [@mattjmcnaughton](https://github.com/mattjmcnaughton))

* Clean up the kube-proxy container image by removing unnecessary packages and files. ([#42090](https://github.com/kubernetes/kubernetes/pull/42090), [@timstclair](https://github.com/timstclair))
* Better compat with very old iptables (e.g. CentOS 6) ([#37594](https://github.com/kubernetes/kubernetes/pull/37594), [@thockin](https://github.com/thockin))

### Scheduler

    * Notes on LB name:
    * to ensure backword and forward compat:
    * - SingleStack -v4 (pre v1.16) => BackendPool name == clusterName
    * - SingleStack -v6 => BackendPool name == clusterName (all cluster bootstrap uses this name)
    * DualStack:

*   Development of Kubernetes Federation has moved to [github.com/kubernetes/federation](https://github.com/kubernetes/federation). This move out of tree also means that Federation will begin releasing separately from Kubernetes. Impact:
    *   Federation-specific behavior will no longer be included in kubectl
    *   kubefed will no longer be released as part of Kubernetes

- Fixed a bug which the scheduler didn't correctly tell plugins Node deletion.

- Performance regression #97685 has been fixed....

  Server-side apply now considers `.spec.defaultBackend` to be an atomic struct.  This means that any field-owner who sets values in that struct (they are mutually exclusive) owns the whole struct. For almost all users this change has no impact; for controllers that want to change the default backend port from number to name (or vice-versa), this makes it easier. ([#126207](https://github.com/kubernetes/kubernetes/pull/126207), [@thockin](https://github.com/thockin)) [SIG API Machinery]...

- The `metadata.clusterName` field is completely removed. This should not have any user-visible impact. ([#109602](https://github.com/kubernetes/kubernetes/pull/109602), [@lavalamp](https://github.com/lavalamp))

that denies admission based at least partially on the old state of the
Node object.

**Note**: This only impacts validating admission plugins that rely on old
values in certain fields, and does not impact calls from kubelets that go
through the built-in NodeRestriction admission plugin.

**Affected Versions**:
  - kube-apiserver v1.20.0 - v1.20.5
  - kube-apiserver v1.19.0 - v1.19.9
  - kube-apiserver <= v1.18.17