// "ClusterIP" allocates a cluster-internal IP address for load-balancing
  // to endpoints. Endpoints are determined by the selector or if that is not
  // specified, by manual construction of an Endpoints object or
  // EndpointSlice objects. If clusterIP is "None", no virtual IP is
  // allocated and the endpoints are published as a set of endpoints rather
  // than a virtual IP.

- One new field "InternalTrafficPolicy" in Service is added.
  It specifies if the cluster internal traffic should be routed to all endpoints or node-local endpoints only.
  "Cluster" routes internal traffic to a Service to all endpoints.
  "Local" routes traffic to node-local endpoints only, and traffic is dropped if no node-local endpoints are ready.

      endpoint now gets broken when that endpoint becomes non-ready
      (rather than continuing until the endpoint is fully deleted).
  
    - Traffic to a service IP now starts getting rejected (as opposed to
      merely dropped) as soon as there are no longer any *usable*
      endpoints, rather than waiting until all of the terminating
      endpoints have terminated even when those terminating endpoints

- Fix kube-apiserver not to create default/kubernetes service endpoints before it reports readiness via the /healthz and therefore is ready to serve requests. Also early during startup old endpoints are remove which might be left over from a previously crashed kube-apiserver. ([#74668](https://github.com/kubernetes/kubernetes/pull/74668), [@sttts](https://github.com/sttts))

- The endpointslice mirroring controller mirrors endpoints annotations and labels to the generated endpoint slices, it also ensures that updates on any of these fields on the endpoints are mirrored. 

      endpoint now gets broken when that endpoint becomes non-ready
      (rather than continuing until the endpoint is fully deleted).
  
    - Traffic to a service IP now starts getting rejected (as opposed to
      merely dropped) as soon as there are no longer any *usable*
      endpoints, rather than waiting until all of the terminating
      endpoints have terminated even when those terminating endpoints

- Fixed a bug where improper storage and comparison of endpoints led to excessive API traffic from the endpoints controller ([#94935](https://github.com/kubernetes/kubernetes/pull/94935), [@damemi](https://github.com/damemi)) [SIG Apps, Network and Testing]

- resource metrics endpoint `/metrics/resource/v1alpha1` as well as all metrics under this endpoint have been deprecated. Please convert to the following metrics emitted by endpoint `/metrics/resource`:
  - scrape_error --> scrape_error
  - node_cpu_usage_seconds_total --> node_cpu_usage_seconds

- Fixing issue with Winkernel Proxier - No ingress load balancer rules with endpoints to support load balancing when all the endpoints are terminating. ([#114451](https://github.com/kubernetes/kubernetes/pull/114451),...

## Changelog since v1.12.7

### Other notable changes

* Connections from Pods to Services with 0 endpoints will now ICMP reject immediately, rather than blackhole and timeout. ([#72534](https://github.com/kubernetes/kubernetes/pull/72534), [@thockin](https://github.com/thockin))