has accidentally broken Calico''s rules. Set to 0 to disable IP
                  sets refresh. Note: the default for this value is lower than the
                  other refresh intervals as a workaround for a Linux kernel bug that
                  was fixed in kernel version 4.11. If you are using v4.11 or greater
                  you may want to set this to, a higher value to reduce Felix CPU

- Systems running `iptables` 1.8.0 or newer should start it in legacy mode. Please note that this affects all versions of Kubernetes and not only v1.16.0. For more detailed information about the issue and how to apply a workaround, please refer to the official documentation
- Generating informers for packages in directories containing dots in their name is broken. This will be fixed in v1.16.1. ([#82860](https://github.com/kubernetes/kubernetes/issues/82860))

  This issue only could happen if you use custom plugins or if you change plugin set being used in your scheduler via the scheduler config. ([#122013](https://github.com/kubernetes/kubernetes/pull/122013), [@sanposhiho](ht...

          DROP, which causes outbound container traffic to stop working by
          default. See
          [#40182](https://github.com/kubernetes/kubernetes/issues/40182) for
          the workaround.

        - The support for the v1 registries was removed.

- Docker 1.12.6

    - Overlay2 and live-restore are not validated.

    - **Known issues**

- [`KT-59677`](https://youtrack.jetbrains.com/issue/KT-59677) K2: Report diagnostics about missing receiver for delegated constructor call to inner class
- [`KT-65183`](https://youtrack.jetbrains.com/issue/KT-65183) K2: Remove workaround for `@OnlyInputTypes` and captured types with recursive supertypes from inference
- [`KT-66120`](https://youtrack.jetbrains.com/issue/KT-66120) IrFakeOverrideBuilder: wrong return type in intersection with 3 classes

- `kubeadm`: generate CA certificates with a start time that is offset 5
  minutes in the past relative to the current system time to workaround cases of
  clock desync. ([#118922](https://github.com/kubernetes/kubernetes/pull/118922), [@champtar](https://github.com/champtar))
- `plugin_evaluation_total` metric supports prescore/score extension point.

                               !mean_type_tensor.hasStaticShape()))
      return failure();

    // TODO(b/69928690): Support mixed precision in the XLA batch
    // normalization operators. As a workaround, create a new x with the same
    // element type as scale (which may be more precise than the input type).
    Value bn_train_input = rewriter.create<mhlo::ConvertOp>(
        op.getLoc(), op.getX(), scale_element_type);

- Kubernetes cannot distinguish between GCE Zonal PDs and Regional PDs with the same name. To workaround this issue, precreate PDs with unique names. PDs that are dynamically provisioned do not encounter this issue. ([#70716](https://github.com/kubernetes/kubernetes/pull/70716), [@msau42](https://github.com/msau42))

## Known Issues

- the IP allocator in v1.17.0 can return errors such as `the cluster IP <ip> for service <service-name>...

	}
	switch service.Spec.Type {
	case core.ServiceTypeLoadBalancer:
		for ix := range service.Spec.Ports {
			port := &service.Spec.Ports[ix]
			// This is a workaround for broken cloud environments that
			// over-open firewalls.  Hopefully it can go away when more clouds
			// understand containers better.
			if port.Port == ports.KubeletPort {