[`tf.distribute.experimental.PreemptionCheckpointHandler`](https://www.tensorflow.org/api_docs/python/tf/distribute/experimental/PreemptionCheckpointHandler)
        to handle worker preemption/maintenance and cluster-wise consistent
        error reporting for `tf.distribute.MultiWorkerMirroredStrategy`.
        Specifically, for the type of interruption with advance notice, it

## Changes since v1.2.2

### Action Required

* Make watch cache treat resourceVersion consistent with uncached watch ([#24008](https://github.com/kubernetes/kubernetes/pull/24008), [@liggitt](https://github.com/liggitt))

### Other notable changes

 *         });
 * }
 *
 * <p>The returned cache implements all optional operations of the {@link LoadingCache} and {@link
 * Cache} interfaces. The {@code asMap} view (and its collection views) have <i>weakly consistent
 * iterators</i>. This means that they are safe for concurrent use, but if other threads modify the
 * cache after the iterator is created, it is undefined which of these changes, if any, are

     * type. So why even make that error possible by giving callers the choice?
     *
     * On the other hand, the current signature is consistent with the similar allAsList method. And
     * eventually this method may go away entirely in favor of an API like
     * whenAllComplete().collectSuccesses(). That API would have a signature more like the current
     * one.

      "Invalid URL host: \"[0.0.0.0:0:0:0:0:0:1]\"",
    )
  }

  @Test
  fun hostIpv6WithIncompleteIpv4Suffix() {
    // To Chrome & Safari these are well-formed; Firefox disagrees. (We're consistent with Firefox).
    assertInvalid(
      "http://[0:0:0:0:0:1:255.255.255.]/",
      "Invalid URL host: \"[0:0:0:0:0:1:255.255.255.]\"",
    )
    assertInvalid(
      "http://[0:0:0:0:0:1:255.255.255]/",

...

resource version option, when passed to a list call, is now consistently interpreted as the minimum allowed resource version.  Previously when listing resources that had the watch cache disabled clients could retrieve a snapshot at that exact resource version.  If the client requests a resource version newer than the current state, a TimeoutError is returned suggesting the client retry in a few seconds.  This behavior is now consistent for both single item retrieval and list calls, and for when the...

		}

		fileName = part.FileName()

		// Multiple values for the same key (one map entry, longer slice) are cheaper
		// than the same number of values for different keys (many map entries), but
		// using a consistent per-value cost for overhead is simpler.
		maxMemoryBytes := 2 * int64(10<<20)
		maxMemoryBytes -= int64(len(name))
		maxMemoryBytes -= mapEntryOverhead
		if maxMemoryBytes < 0 {

- Component-base: make the validation logic about LeaderElectionConfiguration consistent between component-base and client-go ([#111758](https://github.com/kubernetes/kubernetes/pull/111758), [@SataQiu](https://github.com/SataQiu)) [SIG API Machinery and Scheduling]

release, SIG Apps moved the Kubernetes workloads API to the new apps/v1beta2 group and version. The DaemonSet, Deployment, ReplicaSet, and StatefulSet objects are affected by this change. The new apps/v1beta2 group and version provide a stable and consistent API surface for building applications in Kubernetes. For details about deprecations and behavioral changes, see [Notable Features](#notable-features). SIG Apps intends to promote this version to GA in a future release.

[SIG Apps]: ...