- Implemented API streaming for the `watch-cache`
  
  When `sendInitialEvents` `ListOption` is set together with `watch=true`, it begins the watch stream with synthetic init events followed by a synthetic "Bookmark" after which the server continues streaming events. ([#110960](https://github.com/kubernetes/kubernetes/pull/110960), [@p0lyn0mial](https://github.com/p0lyn0mial))
- Introduced API for streaming.
  

* Map correct vmset name for Azure internal load balancers. ([#59747](https://github.com/kubernetes/kubernetes/pull/59747), [@feiskyer](https://github.com/feiskyer))

    * `externalTrafficPolicy` field, the values chnag as follows:
          * "OnlyLocal" becomes "Local"
          * "Global" becomes "Cluster".

- CoreDNS correctly implements DNS spec for Services with externalNames that look like IP addresses. Kube-dns does not follow the spec for the same case, resulting in a behavior change when moving from Kube-dns to CoreDNS. See: [coredns/coredns#2324](https://github.com/coredns/coredns/issues/2324)

- Shell completion is now provided for the "--subresource" flag. ([#109070](https://github.com/kubernetes/kubernetes/pull/109070), [@marckhouzam](https://github.com/marckhouzam))
- Some apiserver metrics were changed, as follows.
  - `priority_level_seat_count_samples` is replaced with `priority_level_seat_utilization`, which samples every nanosecond rather than every millisecond; the old metric conveyed utilization despite its name.

- YAML documents separators ("---") can now be followed by whitespace and comments ("# ....") on the same line. This fixes a bug where documents starting with a comment after the separator were ignored. Other types of content on the same line will result in an error. ([#103457](https://github.com/kub...

- Modify DNS-1123 error messages to indicate that RFC 1123 is not followed exactly ([#94182](https://github.com/kubernetes/kubernetes/pull/94182), [@mattfenwick](https://github.com/mattfenwick)) [SIG API Machinery, Apps, Auth, Network and Node]

- The kube-apiserver's Prometheus metrics have been extended with some that describe the costs of handling LIST requests.  They are as follows.
  - *apiserver_cache_list_total*: Counter of LIST requests served from watch cache, broken down by resource_prefix and index_name

- Rename unschedulableQ to unschedulablePods ([#108919](https://github.com/kubernetes/kubernetes/pull/108919), [@denkensk](https://github.com/denkensk))
- SPDY transport in client-go will no longer follow redirects. ([#108531](https://github.com/kubernetes/kubernetes/pull/108531), [@tallclair](https://github.com/tallclair))

## Known Issues

- Concurrently joining control-plane nodes does not work as expected in kubeadm 1.15.0. The feature was planned for release in 1.15.0, but a fix may  come in a follow up patch release.