Kube-proxy's IPVS proxy mode no longer sets the net.ipv4.conf.all.route_localnet sysctl parameter. Nodes upgrading will have net.ipv4.conf.all.route_localnet set to 1 but new nodes will inherit the system default (usually 0). If you relied on any behavior requiring net.ipv4.conf.all.route_localnet, you must set ensure it is enabled as kube-proxy will no longer set it automatically. This change helps to further mitigate CVE-2020-8558. ([#92938](https://github.com/kubernetes/kubernetes/pull/92938), [@lbernai...

  [overwrite of manual edits to GCP Health Checks](https://github.com/kubernetes/ingress/issues/842)
  managed by the GLBC Ingress Controller. This can cause the health checks to start failing,
  requiring you to reapply the manual edits.
  * This issue does not affect clusters that were already running Kubernetes v1.6.4 or higher.
  * This issue does not affect Health Checks that were left in the configuration

- Removed an inaccurate statement about requiring ports when the Pod spec `hostNetwork` field was set. ([#130994](https://github.com/kubernetes/kubernetes/pull/130994), [@BenTheElder](https://github.com/BenTheElder)) [SIG Network and Node]

- In "large" clusters, kube-proxy in iptables mode will now sometimes
  leave unused rules in iptables for a while (up to `--iptables-sync-period`)
  before deleting them. This improves performance by not requiring it to
  check for stale rules on every sync. (In smaller clusters, it will still
  remove unused rules immediately once they are no longer used.)

  (The threshold for "large" used here is currently "1000 endpoints" but

feature to support referencing counter sets across ResourceSlices within the same resource pool. Devices from incomplete pools were no longer considered for allocation. This change introduced backwards-incompatible updates to the alpha feature, requiring any ResourceSlices using it to be removed before upgrading or downgrading between v1.34 and v1.35. ([#134189](https://github.com/kubernetes/kubernetes/pull/134189), [@mortent](https://github.com/mortent)) [SIG API Machinery, Node, Scheduling and...

### Uncategorized

- Deprecate kubectl version long output, will be replaced with kubectl version ` --short`. Users requiring full output should use `--output=yaml|json` instead. ([#108987](https://github.com/kubernetes/kubernetes/pull/108987), [@soltysh](https://github.com/soltysh))


## Dependencies

### Added

* [alpha] Improve Flex volume deployment

  * Flex volume driver deployment is simplified in the following ways:

    * New driver files can now be automatically discovered and initialized without requiring a kubelet or controller-manager restart.

    * A sample DaemonSet to deploy Flexvolume drivers is now available.

* [prototype] Volume snapshots

the KUBECONFIG file.) This way, users can now  set a Google Service Account JSON key in the  GOOGLE_APPLICATION_CREDENTIALS environment variable, craft a kubeconfig file with GKE master IP+CA cert, and authenticate to GKE in headless mode without requiring gcloud CLI. ([#58141](https://github.com/kubernetes/kubernetes/pull/58141), [@ahmetb](https://github.com/ahmetb))

* kubectl port-forward no longer supports the deprecated -p <pod-name> flag; the flag itself is unnecessary and should...

     * the lock to check again. Since an IDLE state was observed inside the preceding synchronized
     * block, and reference field assignment is atomic, this may save reacquiring the lock when
     * another thread or the worker task has cleared the count and set the state.
     *
     * <p>When {@link #executor} is a directExecutor(), the value written to

   * "pass on" the signal to another thread. If it *was* signalled, then its guard must have been
   * satisfied at the time of signal, and has since been modified by some other thread to be
   * non-satisfied before reacquiring the lock, and that other thread takes over the responsibility
   * of signaling the next waiter.
   *
   * Unlike the underlying Condition, if we are not careful, an interrupt *can* cause a signal to be