registers all of its command-line flags with an internal flagset, which prevents flags from third party libraries from unintentionally leaking into the Kubelet's command-line API. Many unintentionally leaked flags are now marked deprecated, so that users have a chance to migrate away from them before they are removed. In addition, one previously leaked flag, --cloud-provider-gce-lb-src-cidrs, has been entirely removed from the Kubelet's command-line API, because it is irrelevant to Kubelet operation. The...

Some upgrade methods might upgrade etcd from 3.0 to 3.1 automatically when you upgrade from Kubernetes 1.8, unless you specify otherwise. Because [etcd does not support downgrading](https://coreos.com/etcd/docs/latest/upgrades/upgrade_3_1.html), you'll need to either remain on etcd 3.1 or restore from a backup if you want to downgrade back to Kubernetes 1.8.

## Introduction to 1.9.0

- Users will see increase in time for deletion of pods and also guarantee that removal of pod from api server  would mean deletion of all the resources from container runtime. ([#92817](https://github.com/kubernetes/kubernetes/pull/92817), [@kmala](https://github.com/kmala)) [SIG Node]

    * cAdvisor metrics can now be scraped from `/metrics/cadvisor` on the kubelet ports.
    * Note that you have to update your scraping jobs to get kubelet-only metrics from `/metrics` and `container_*` metrics from `/metrics/cadvisor`

- Kube-apiserver: fixes a 1.27+ regression in watch stability by serving watch requests without a resourceVersion from the watch cache by default, as in <1.27 (disabling the change in #115096 by default). This mitigates the impact of an etcd watch bug (https://github.com/etcd-io/etcd/pull/17555). If the 1.27 change in #115096 to serve these requests from underlying storage is still desired despite the impact on watch stability, it can be re-enabled with a `WatchFromStorageWithoutResourceVersion`...

- Kube-apiserver: fixes a 1.27+ regression in watch stability by serving watch requests without a resourceVersion from the watch cache by default, as in <1.27 (disabling the change in #115096 by default). This mitigates the impact of an etcd watch bug (https://github.com/etcd-io/etcd/pull/17555). If the 1.27 change in #115096 to serve these requests from underlying storage is still desired despite the impact on watch stability, it can be re-enabled with a `WatchFromStorageWithoutResourceVersion`...

- Kubeadm: get k8s CI version markers from k8s infra bucket ([#98836](https://github.com/kubernetes/kubernetes/pull/98836), [@hasheddan](https://github.com/hasheddan)) [SIG Cluster Lifecycle and Release]
- Kubelet should ignore cgroup driver check on Windows node. ([#98384](https://github.com/kubernetes/kubernetes/pull/98384), [@pacoxu](https://github.com/pacoxu)) [SIG Node]
- TerminationGracePeriodSeconds from pod spec is respected for the mirror pod

- kube-controller-manager
  - The deprecated `--insecure-experimental-approve-all-kubelet-csrs-for-group` flag has been removed.
- kubelet

  - *apiserver_cache_list_fetched_objects_total*: Counter of objects read from watch cache in the course of serving a LIST request, broken down by resource_prefix and index_name
  - *apiserver_cache_list_evaluated_objects_total*: Counter of objects tested in the course of serving a LIST request from watch cache, broken down by resource_prefix

  - apiserver-advertise-address / apiserver-bind-port: use either localAPIEndpoint from InitConfiguration or controlPlaneEndpoint from ClusterConfiguration.
  - cluster-name: use clusterName from ClusterConfiguration
  - cert-dir: use certificatesDir from ClusterConfiguration ([#94879](https://github.com/kubernetes/kubernetes/pull/94879), [@knight42](https://github.com/knight42)) [SIG Cluster Lifecycle]