* Free up CPU and memory requested but unused by Metrics Server Pod Nanny. ([#57252](https://github.com/kubernetes/kubernetes/pull/57252), [@kawych](https://github.com/kawych))
* Configurable liveness probe initial delays for etcd and kube-apiserver in GCE ([#57749](https://github.com/kubernetes/kubernetes/pull/57749), [@wojtek-t](https://github.com/wojtek-t))

*   `tf.data`

    * `tf.data` now has an `autotune_options.initial_parallelism` option to
      control the initial parallelism setting used by autotune before the data
      pipeline has started running. The default is 16. A lower value reduces
      initial memory usage, while a higher value improves startup time.

## Keras

*  `keras.layers.experimental.DynamicEmbedding`

.gy-xxl-4 {\n    --bs-gutter-y: 1.5rem;\n  }\n  .g-xxl-5,\n  .gx-xxl-5 {\n    --bs-gutter-x: 3rem;\n  }\n  .g-xxl-5,\n  .gy-xxl-5 {\n    --bs-gutter-y: 3rem;\n  }\n}\n.table {\n  --bs-table-color-type: initial;\n  --bs-table-bg-type: initial;\n  --bs-table-color-state: initial;\n  --bs-table-bg-state: initial;\n  --bs-table-color: var(--bs-body-color);\n  --bs-table-bg: var(--bs-body-bg);\n  --bs-table-border-color: var(--bs-border-color);\n  --bs-table-accent-bg: transparent;\n  --bs-table-striped-color:...

   * avoid memory leaks when an element is no longer necessary.
   *
   * <p>This method accepts an array with element type {@code @Nullable T}, but callers must pass an
   * array whose contents are initially non-null. The {@code @Nullable} annotation indicates that
   * this method will write nulls into the array during iteration.
   *
   * <p>This is mainly just to avoid the intermediate ArrayDeque in ConsumingQueueIterator.

    value has always been nullable but it wasn't declared properly.
 *  Fix: Reverse suppression of connect exceptions. When both a call and its retry fail, we now
    throw the initial exception which is most likely to be actionable.
 *  Fix: Retain interrupted state when throwing `InterruptedIOException`. A single interrupt should
    now be sufficient to break out an in-flight OkHttp call.

// -------------------------
// This operation lists in-progress multipart uploads. An in-progress
// multipart upload is a multipart upload that has been initiated,
// using the Initiate Multipart Upload request, but has not yet been
// completed or aborted. This operation returns at most 1,000 multipart
// uploads in the response.

* Added configurable etcd initial-cluster-state to kube-up script ([#41320](https://github.com/kubernetes/kubernetes/pull/41320), [@jszczepkowski](https://github.com/jszczepkowski))

		HTTPStatusCode: http.StatusBadRequest,
	},
	ErrSSEMultipartEncrypted: {
		Code:           "InvalidRequest",
		Description:    "The multipart upload initiate requested encryption. Subsequent part requests must include the appropriate encryption parameters.",
		HTTPStatusCode: http.StatusBadRequest,
	},
	ErrSSEEncryptedObject: {
		Code:           "InvalidRequest",

- Fix: don't use docker config cache if it's empty ([#92330](https://github.com/kubernetes/kubernetes/pull/92330), [@andyzhangx](https://github.com/andyzhangx)) [SIG Cloud Provider]
- Fix: initial delay in mounting azure disk & file ([#93052](https://github.com/kubernetes/kubernetes/pull/93052), [@andyzhangx](https://github.com/andyzhangx)) [SIG Cloud Provider and Storage]

        .build(),
    )
    queueDispatcher.enqueue(MockResponse(body = "call2 response"))
    queueDispatcher.enqueue(MockResponse(body = "call1 response"))

    // We use a re-entrant dispatcher to initiate one HTTPS connection while the other is in flight.
    server.dispatcher =
      object : Dispatcher() {
        var requestCount = 0

        override fun dispatch(request: RecordedRequest): MockResponse {