### Bug or Regression

- Fixes a race condition in the iptables mode of kube-proxy in 1.27 and later
  that could result in some updates getting lost (e.g., when a service gets a
  new endpoint, the rules for the new endpoint might not be added until
  much later). ([#122757](https://github.com/kubernetes/kubernetes/pull/122757), [@hakman](https://github.com/hakman)) [SIG Network]

### Bug or Regression

- Fixes a race condition in the iptables mode of kube-proxy in 1.27 and later
  that could result in some updates getting lost (e.g., when a service gets a
  new endpoint, the rules for the new endpoint might not be added until
  much later). ([#122815](https://github.com/kubernetes/kubernetes/pull/122815), [@wedaly](https://github.com/wedaly)) [SIG Network]

Added a preflight check `ContainerRuntimeVersion` to validate if the installed container runtime supports the `RuntimeConfig` gRPC method. If unsupported, `kubeadm` prints a warning message.
  
  Starting with Kubernetes `v1.36`, `kubelet` might refuse to start if the CRI runtime does not support this feature. More information can be found at the [Kubernetes blog](https://kubernetes.io/blog/2025/09/12/kubernetes-v1-34-cri-cgroup-driver-lookup-now-ga/). ([#134906](https://github.com/kubern...

	// FIXME: network calls made in this manner such as one goroutine per node,
	// can easily eat into the internode bandwidth. This function would be mostly
	// TX saturating, however there are situations where a RX might also saturate.
	// To avoid these problems we must split the work at scale. With 1000 node
	// setup becoming a reality we must try to shard the work properly such as

 - CSINode initialization does not crash kubelet on startup when APIServer is not reachable or kubelet has not the right credentials yet. ([#89589](https://github.com/kubernetes/kubernetes/pull/89589), [@jsafrane](https://github.com/jsafrane)) [SIG Storage]

	}

	return subSys, inputs, tgt, e
}

// kvFields - converts an input string of form "k1=v1 k2=v2" into
// fields of ["k1=v1", "k2=v2"], the tokenization of each `k=v`
// happens with the right number of input keys, if keys
// input is empty returned value is empty slice as well.
func kvFields(input string, keys []string) []string {
	valueIndexes := make([]int, 0, len(keys))
	for _, key := range keys {

  which the scheduler might miss some updates for the Pods rejected by NodeUnschedulable plugin and put the Pods in the queue for a longer time than needed. ([#127427](https://github.com/kubernetes/kubernetes/pull/127427), [@sanposhiho](https://github.com/sanposhiho)) [SIG Scheduling]
- Fixes the bug in PodTopologySpread that only happens with QHint enabled, 

but in order to avoid inconsistency the Agreement is copyrighted and
may only be modified in the following manner. The Agreement Steward
reserves the right to publish new versions (including revisions) of
this Agreement from time to time. No one other than the Agreement
Steward has the right to modify this Agreement. The Eclipse Foundation
is the initial Agreement Steward. The Eclipse Foundation may assign the

   ([#121461](https://github.com/kubernetes/kubernetes/pull/121461), [@alexzielenski](https://github.com/alexzielenski))
- Scheduler now skips the `NodeAffinity Score` plugin when it has nothing to do with a Pod. You might have noticed an increase in the metric `plugin_execution_duration_seconds` for `extension_point=score` and `plugin=NodeAffinity`, because the plugin only runs when it's relevant.

        .build(),
    )

    // At least three request/response pairs are required because after the first request is cached
    // a different execution path might be taken. Thus modifications to the cache applied during
    // the second request might not be visible until another request is performed.
    assertThat(get(server.url("/")).body.string()).isEqualTo("ABCABCABC")