- The e2e suite can be instructed not to wait for pods in kube-system to be ready or for all nodes to be ready by passing `--allowed-not-ready-nodes=-1` when invoking the e2e.test program. This allows callers to run subsets of the e2e suite in scenarios other than perfectly healthy clusters. ([#98781](https://github.com/kubernetes/kubernetes/pull/98781), [@...

- Containers which specify a `startupProbe` but not a `readinessProbe` were previously considered "ready" before the `startupProbe` completed, but are now considered "not-ready". ([#92196](https://github.com/kubernetes/kubernetes/pull/92196), [@thockin](https://github.com/thockin)) [SIG Node]

*   Taint key `unreachable` is now in GA.
*   Taint key `notReady` is changed to `not-ready`, and is also now in GA.
*   These changes are automatically updated for taints. Tolerations for these taints must be updated manually. Specifically, you must:
    *   Change `node.alpha.kubernetes.io/notReady` to `node.kubernetes.io/not-ready`
    *   Change `node.alpha.kubernetes.io/unreachable` to  `node.kubernetes.io/unreachable`

* Retry 'connection refused' errors when setting up clusters on GCE. ([#57394](https://github.com/kubernetes/kubernetes/pull/57394), [@mborsz](https://github.com/mborsz))

* YAMLDecoder Read now returns the number of bytes read ([#57000](https://github.com/kubernetes/kubernetes/pull/57000), [@sel](https://github.com/sel))

  - The deprecated `--etcd-quorum-read` flag has been removed. Quorum reads are now always enabled when fetching data from etcd. Remove the `--etcd-quorum-read` flag from kube-apiserver invocations before upgrading.
- kube-controller-manager

    - [Node binaries](#node-binaries-7)
  - [Changelog since v1.17.8](#changelog-since-v1178)
  - [Urgent Upgrade Notes](#urgent-upgrade-notes)
    - [(No, really, you MUST read this before you upgrade)](#no-really-you-must-read-this-before-you-upgrade)
  - [Changes by Kind](#changes-by-kind-8)
    - [Bug or Regression](#bug-or-regression-7)
  - [Dependencies](#dependencies-8)
    - [Added](#added-8)

    * To avoid that negative effect only use per pod metrics from pods that are:
    * - ready (so metrics about them should be valid), or
    * - unready but creation and last readiness change timestamps are apart more than 10s (pods that have formerly been ready and so metrics are in at least some cases (pod becoming unready because of overload) very useful).

  - [Urgent Upgrade Notes](#urgent-upgrade-notes)
    - [(No, really, you MUST read this before you upgrade)](#no-really-you-must-read-this-before-you-upgrade)
  - [Changes by Kind](#changes-by-kind-13)
    - [Deprecation](#deprecation)
    - [API Change](#api-change-4)
    - [Feature](#feature-12)

* StatefulSets now include an alpha scaling feature accessible by setting the `spec.podManagementPolicy` field to `Parallel`.  The controller will not wait for pods to be ready before adding the other pods, and will replace deleted pods as needed.  Since parallel scaling creates pods out of order, you cannot depend on predictable membership changes within your set. ([#44899](https://github.com/kubernetes/kubernetes/pull/44899),...

* Kubernetes will try to read the openstack auth config from the client config and fall back to read from the environment variables if the auth config is not available. ([#60200](https://github.com/kubernetes/kubernetes/pull/60200), [@dixudx](https://github.com/dixudx))

### SIG Scheduling