- Fixed a bug in kubelet that will saturate CPU utilization after containerd got restarted. ([#97174](https://github.com/kubernetes/kubernetes/pull/97174), [@hanlins](https://github.com/hanlins)) [SIG Node]
- Fixed a bug that causes smaller number of conntrack-max being used under CPU static policy. (#99225, @xh4n3) ([#99613](https://github.com/kubernetes/kubernetes/pull/99613), [@xh4n3](https://github.com/xh4n3)) [SIG Network]

to `true` (it is disabled by default). An API server fronted by an L7 load balancer that already mitigates these http/2 attacks may choose not to enable the kube-apiserver mitigation to avoid disrupting load balancer → kube-apiserver connections if http/2 requests from multiple clients share the same backend connection. An API server on a private network may choose not to enable the kube-apiserver mitigation to prevent performance regressions for unauthenticated clients. Authenticated requests rely...

- Service Controller: will now update load balancer hosts after node's `ProviderID` is
  updated. ([#120492](https://github.com/kubernetes/kubernetes/pull/120492), [@cezarygerard](https://github.com/cezarygerard))

- Apiserver add a new flag --goaway-chance which is the fraction of requests that will be closed gracefully(GOAWAY) to prevent HTTP/2 clients from getting stuck on a single apiserver. 
  After the connection closed(received GOAWAY), the client's other in-flight requests won't be affected, and the client will reconnect. 

parameter in kubeadm has been updated to drop an unnecessary redirection from ci/latest.txt to ci-cross/latest.txt. Users should know exactly where the builds are stored on Google Cloud storage buckets from now on. For example for 1.9 and 1.10, users can specify ci/latest-1.9 and ci/latest-1.10 as the CI build jobs what build images correctly updates those. The CI jobs for master update the ci-cross/latest location, so if you are looking for latest master builds, then the correct parameter to use...

  - **apiserver_flowcontrol_priority_level_seat_count_watermarks**: histograms of high and low watermarks of number of seats occupied by executing requests (both regular and final-delay phases included), broken down by priority_level.

	if c.Empty() {
		return r
	}

	// Now overlay existing object replication choices for target
	if oi.DeleteMarker {
		opts := replication.ObjectOpts{
			Name:           oi.Name,
			DeleteMarker:   oi.DeleteMarker,
			VersionID:      oi.VersionID,

provided status of etcd, kube-scheduler, and kube-controller-manager components, but only worked when those components were local to the API server, and when kube-scheduler and kube-controller-manager exposed unsecured health endpoints. Instead of this API, etcd health is included in the kube-apiserver health check and kube-scheduler/kube-controller-manager health checks can be made directly against those components' health endpoints. ([#93570](https://github.com/kubernetes/kubernetes/pull/93570), [@...

- The pod phase lifecycle guarantees that terminal Pods, those whose states are Unready or Succeeded, can not regress and will have all container stopped. Hence, terminal Pods will never be reachable and should not publish their IP addresses on the Endpoints or EndpointSlices, independently of the Service TolerateUnready option. ([#110258](https://github.com/kubernetes/kubernetes/pull/110258), [@robscott](h...

to `true` (it is disabled by default). An API server fronted by an L7 load balancer that already mitigates these http/2 attacks may choose not to enable the kube-apiserver mitigation to avoid disrupting load balancer → kube-apiserver connections if http/2 requests from multiple clients share the same backend connection. An API server on a private network may choose not to enable the kube-apiserver mitigation to prevent performance regressions for unauthenticated clients. Authenticated requests rely...