SavedModel.
    *   The user object metadata field in the SavedModel proto has been
        deprecated as part of the updates to Keras SavedModel. Keras was the
        only consumer of this field prior to the update.
*   GPU
    *   TF 2.3 includes PTX kernels only for
        [compute capability](https://developer.nvidia.com/cuda-gpus) 7.0 to

option go_package = "k8s.io/apimachinery/pkg/util/intstr";

// IntOrString is a type that can hold an int32 or a string.  When used in
// JSON or YAML marshalling and unmarshalling, it produces or consumes the
// inner type.  This allows you to have, for example, a JSON field that can
// accept a name or number.
// TODO: Rename to Int32OrString
//
// +protobuf=true
// +protobuf.options.(gogoproto.goproto_stringer)=false

* Fixes a flaw (CVE-2019-11253) in json/yaml decoding where large or malformed documents could consume excessive server resources. Request bodies for normal API requests (create/delete/update/patch operations of regular resources) are now limited to 3MB. ([#83261](https://github.com/kubernetes/kubernetes/pull/83261), [@liggitt](https://github.com/liggitt))...

Defaulting will be available as alpha in Kubernetes 1.15 and requires structural schemas.

#### CustomResourceDefinition OpenAPI Publishing

OpenAPI specs for native types have long been served at /openapi/v2, and they are consumed by a number of components, notably kubectl client-side validation, kubectl explain and OpenAPI based client generators.

 *   classes from the main source set,
 * - the test fixtures are automatically made available to the test classpath.
 *
 * Configures the Project as a test fixtures consumer according to the `testFixtures` extension configuration.
 */
plugins {
    `java-library`
    `java-test-fixtures`
    groovy
    id("gradlebuild.dependency-modules")
}

System.out.println("REQUEST 1 (new connection)");
try (Response response = client.newCall(request).execute()) {
  // Consume and discard the response body.
  response.body().source().readByteString();
}

System.out.println("REQUEST 2 (pooled connection)");
try (Response response = client.newCall(request).execute()) {
  // Consume and discard the response body.
  response.body().source().readByteString();
}
```

   * proceed to read the response body with [Response.body]. The response is still live until its
   * response body is [closed][ResponseBody]. The recipient of the callback may consume the response
   * body on another thread.
   *
   * Note that transport-layer success (receiving a HTTP response code, headers and body) does not

    ikrt["Istio krt.Collection"]
    ikrt--"Builds on"-->ikc--"Builds on"-->ic--"Builds on"-->kcg
```

**`kube.Client`** is a fairly light wrapper around Kubernetes clients, but offers quite a few benefits.
Istio consumes _many_ different types of clients (at the time of writing, 7), as well as informers for each of these.
Additionally, we often need to convert to various different client forms for different libraries we integrate with.

    // appendCodePoint(). The Punycode algorithm processes code points in increasing code-point
    // order, not in increasing index order.
    val codePoints = mutableListOf<Int>()

    // consume all code points before the last delimiter (if there is one)
    //  and copy them to output, fail on any non-basic code point
    val lastDelimiter = string.lastIndexOf('-', limit)
    if (lastDelimiter >= pos) {

- Reduced server startup time. For IAM encryption with the root credentials, MinIO had
   to use a memory-hard function (Argon2) that (on purpose) consumes a lot of memory and CPU.
   The new KMS-based approach can use a key derivation function that is orders of magnitudes
   cheaper w.r.t. memory and CPU.