- [SIG VMWare](#sig-vmware-1)
    - [SIG Windows](#sig-windows-1)
  - [Other Notable Changes](#other-notable-changes-11)
    - [Bug Fixes](#bug-fixes)
    - [Not Very Notable (that is, non-user-facing)](#not-very-notable-that-is-non-user-facing)
  - [External Dependencies](#external-dependencies)
- [v1.12.0-rc.2](#v1120-rc2)
  - [Downloads for v1.12.0-rc.2](#downloads-for-v1120-rc2)

* Change kube-proxy & fluentd CPU request to 20m/80m. ([#23646](https://github.com/kubernetes/kubernetes/pull/23646), [@cjcullen](https://github.com/cjcullen))
* Create a new Deployment in kube-system for every version. ([#23512](https://github.com/kubernetes/kubernetes/pull/23512), [@Q-Lee](https://github.com/Q-Lee))

    Previously if you configured a ping interval that would cause OkHttp to send
    pings, but it did not track whether the reply pongs were received. With this
    update OkHttp requires that every ping receive a response: if it does not
    the connection will be closed and the listener's `onFailure()` method will
    be called.

    Web sockets have always been had pings, but pings on HTTP/2 connections is

### GCP

- Fixed liveness probe in fluentd-gcp cluster addon ([#74522](https://github.com/kubernetes/kubernetes/pull/74522), [@Pluies](https://github.com/Pluies))
- Reduced GCE log rotation check from 1 hour to every 5 minutes.  Rotation policy is unchanged (new day starts, log file size > 100MB). ([#72062](https://github.com/kubernetes/kubernetes/pull/72062), [@jpbetz](https://github.com/jpbetz))

### Network

- Added a mechanism that calculates a digest of etcd and the watch cache every 5 minutes and exposes it as the `apiserver_storage_digest` metric. ([#130475](https://github.com/kubernetes/kubernetes/pull/130475), [@serathius](https://github.com/serathius)) [SIG API Machinery, Instrumentation and Testing]

		if v == nil {
			v = set.CreateStringSet(group)
		} else {
			v.Add(group)
		}
		c.iamUserGroupMemberships[member] = v
	}
}

// removeGroupFromMembershipsMap - removes the group from every member
// in the cache. IMPORTANT: Assumes c.Lock() is held by caller.
func (c *iamCache) removeGroupFromMembershipsMap(group string) {
	for member, groups := range c.iamUserGroupMemberships {
		if !groups.Contains(group) {

		poolsArgs := re.ReplaceAllString(cmdLine, `$3`)
		var anonPools []string

		if !strings.Contains(poolsArgs, "{") || !strings.Contains(poolsArgs, "}") {
			// No ellipses pattern. Anonymize host name from every pool arg
			pools := strings.Fields(poolsArgs)
			anonPools = make([]string, len(pools))
			for index, arg := range pools {
				anonPools[index] = anonAddr(arg)
			}

  } else {
    return ToOperation(iter->second);
  }
}

TF_Operation* TF_GraphNextOperation(TF_Graph* graph, size_t* pos) {
  if (*pos == 0) {
    // Advance past the first sentinel nodes in every graph (the source & sink).
    *pos += 2;
  } else {
    // Advance to the next node.
    *pos += 1;
  }

  mutex_lock l(graph->mu);

- For raw block CSI volumes on Kubernetes, kubelet was incorrectly calling CSI NodeStageVolume for every single "map" (i.e. raw block "mount") operation for a volume already attached to the node. This PR ensures it is only called once per volume per node. ([#112403](https://github.com/kubernetes/kubernetes/pull/112403), [@akankshakumari393](https://g...

  }

  /**
   * We've had a bug where we forget the HTTP response when we see response code 401. This causes a
   * new HTTP request to be issued for every call into the URLConnection.
   */
  @Test
  fun unauthorizedResponseHandling() {
    val mockResponse =
      MockResponse(
        code = HttpURLConnection.HTTP_UNAUTHORIZED,