performing a GET then PUT, the kube-apiserver must write the canonical representation of the object to etcd if the current value does not match. That allows external agents to migrate content in etcd from one API version to another, across different storage types, or across varying encryption levels. This fixes a bug introduced in 1.5 where we unintentionally stopped writing the newest data. ([#48394](https://github.com/kubernetes/kubernetes/pull/48394), [@smarterclayton](https://github.com/smarterclayton))...

  
  Multi vCenter support is deprecated as of v1.21. If you have a Kubernetes cluster spanning across multiple vCenter servers, please consider moving all k8s nodes to a single vCenter Server. vSphere CSI Driver does not support Kubernetes deployment spanning across multiple vCenter servers.
  

For initial installations, see the [Setup topics](https://kubernetes.io/docs/setup/pick-right-solution/) in the Kubernetes documentation.

To upgrade to this release from a previous version, first take any actions required [Before Upgrading](https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG/CHANGELOG-1.9.md#before-upgrading).

func.func @convert_reduce_to_max_int(%arg0: tensor<1x4xi32>) -> tensor<1xi32> {
  // -2147483648 is MIN for INT32
  %0 = mhlo.constant dense<-2147483648> : tensor<i32>
  %1 = mhlo.reduce(%arg0 init: %0) across dimensions = [1] : (tensor<1x4xi32>, tensor<i32>) -> tensor<1xi32>
   reducer(%arg2: tensor<i32>, %arg3: tensor<i32>)  {
    %892 = mhlo.maximum %arg2, %arg3 : tensor<i32>
    "mhlo.return"(%892) : (tensor<i32>) -> ()
  }

performing a GET then PUT, the kube-apiserver must write the canonical representation of the object to etcd if the current value does not match. That allows external agents to migrate content in etcd from one API version to another, across different storage types, or across varying encryption levels. This fixes a bug introduced in 1.5 where we unintentionally stopped writing the newest data. ([#48394](https://github.com/kubernetes/kubernetes/pull/48394), [@smarterclayton](https://github.com/smarterclayton))...

	{VSUM2SWS, 0xfc0007ff00000000, 0x1000068800000000, 0x0, // Vector Sum across Half Signed Word Saturate VX-form (vsum2sws VRT,VRA,VRB)
		[6]*argField{ap_VecReg_6_10, ap_VecReg_11_15, ap_VecReg_16_20}},
	{VSUM4SBS, 0xfc0007ff00000000, 0x1000070800000000, 0x0, // Vector Sum across Quarter Signed Byte Saturate VX-form (vsum4sbs VRT,VRA,VRB)
		[6]*argField{ap_VecReg_6_10, ap_VecReg_11_15, ap_VecReg_16_20}},

- For volumes that allow attaches across multiple nodes, attach and detach operations across different nodes are now executed in parallel. ([#88678](https://github.com/kubernetes/kubernetes/pull/88678), [@verult](https://github.com/verult)) [SIG Storage]

A security issue was discovered in kube-apiserver that could allow an attacker controlled aggregated API server to redirect client traffic to any URL.  This could lead to the client performing unexpected actions as well as leaking the client's credentials to third parties. 

New features include improved integration between the Keystone service and Kubernetes RBAC, and a number of stability and compatibility improvements across the entire provider code-base.

### SIG Scheduling

- Add a new `distribute-cpus-across-numa` option to the static `CPUManager` policy. When enabled, this will trigger the `CPUManager` to evenly distribute CPUs across NUMA nodes in cases where more than one NUMA node is required to satisfy the allocation. ([#105631](https://github.com/kubernetes/kubernetes/pull/105631), [@klueska](https://github.com/klueska))...