- Kube-proxy uses  field-selector clusterIP!=None on Services to avoid watching for Headless Services, reduce  unnecessary network bandwidth ([#126769](https://github.com/kubernetes/kubernetes/pull/126769), [@Sakuralbj](https://github.com/Sakuralbj)) [SIG Network]

- 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 the new object does not specify them. ([#104673](https://github.com/kubernetes/kubernetes/pull/104673), [@thockin](https:...

      * if the new state is "normal" we resume normal operation (go back to default limiter settings),
      * if all zones in the cluster are in "fullDisruption" state we stop all evictions.
* Add a flag for `kubectl expose`to set ClusterIP and allow headless services ([#28239](https://github.com/kubernetes/kubernetes/pull/28239), [@ApsOps](https://github.com/ApsOps))

* Bump up addon kube-dns to v20 for graceful termination ([#33774](https://github.com/kubernetes/kubernetes/pull/33774), [@MrHohn](https://github.com/MrHohn))
* Creating LoadBalancer Service with "None" ClusterIP is no longer possible ([#33274](https://github.com/kubernetes/kubernetes/pull/33274), [@nebril](https://github.com/nebril))

- Updated the API documentation for Service.Spec.IPFamily to warn that its exact
  semantics will probably change before the dual-stack feature goes GA, and users
  should look at ClusterIP or Endpoints, not IPFamily, to figure out if an existing
  Service is IPv4, IPv6, or dual-stack. ([#91527](https://github.com/kubernetes/kubernetes/pull/91527), [@danwinship](https://github.com/danwinship)) [SIG Apps and Network]

- Fixes a bug where Services using finalizers may hold onto ClusterIP and/or NodePort allocated resources for longer than expected if the finalizer is removed using the status subresource ([#120657](https://github.com/kubernetes/kubernetes/pull/120657), [@aojea](https://github.com/aojea)) [SIG Network and Testing]...

- Fixes a bug where Services using finalizers may hold onto ClusterIP and/or NodePort allocated resources for longer than expected if the finalizer is removed using the status subresource ([#120656](https://github.com/kubernetes/kubernetes/pull/120656), [@aojea](https://github.com/aojea)) [SIG Network and Testing]...

- 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 the new object does not specify them. ([#104672](https://github.com/kubernetes/kubernetes/pull/104672), [@thockin](https:...

* Properly manage security groups for loadbalancer services on OpenStack. ([#65373](https://github.com/kubernetes/kubernetes/pull/65373), [@multi-io](https://github.com/multi-io))
* Allow access to ClusterIP from the host network namespace when kube-proxy is started in IPVS mode without either masqueradeAll or clusterCIDR flags ([#65388](https://github.com/kubernetes/kubernetes/pull/65388), [@lbernail](https://github.com/lbernail))

* Add more conditions to the list of predicate failures that won't be resolved by preemption. ([#64995](https://github.com/kubernetes/kubernetes/pull/64995), [@bsalamat](https://github.com/bsalamat))
* Allow access to ClusterIP from the host network namespace when kube-proxy is started in IPVS mode without either masqueradeAll or clusterCIDR flags ([#65388](https://github.com/kubernetes/kubernetes/pull/65388), [@lbernail](https://github.com/lbernail))