*
     * <p><b>Performance note:</b> {@link ArrayList} and {@link java.util.ArrayDeque} consistently
     * outperform {@code LinkedList} except in certain rare and specific situations. Unless you have
     * spent a lot of time benchmarking your specific needs, use one of those instead. (However, we
     * do not currently offer a {@link Multimap} implementation based on {@link
     * java.util.ArrayDeque}.)

 *  New: We've added `requestFailed()` and `responseFailed()` methods to `EventListener`. These
    are called instead of `requestBodyEnd()` and `responseBodyEnd()` in some failure situations.
    They may also be fired in cases where no event was published previously. In this release we did
    an internal rewrite of our event code to fix problems where events were lost or unbalanced.

    when a request body needs to cancel its own call.


## Version 4.2.1

_2019-10-02_

 *  Fix: In 4.1.0 we introduced a performance regression that prevented connections from being
    pooled in certain situations. We have good test coverage for connection pooling but we missed
    this because it only occurs if you have proxy configured and you share a connection pool among
    multiple `OkHttpClient` instances.

    logs.

 *  **WebSocket API now uses `RequestBody` and `ResponseBody` for messages.**
    This is a backwards-incompatible API change.

 *  **The DNS service is now pluggable.** In some situations this may be useful
    to manually prioritize specific IP addresses.

 *  Fix: Don't throw when converting an `HttpUrl` to a `java.net.URI`.
    Previously URLs with special characters like `|` and `[` would break when

* vSphere Volume Plugin Implementation ([#24947](https://github.com/kubernetes/kubernetes/pull/24947), [@abithap](https://github.com/abithap))
* ResourceQuota controller uses rate limiter to prevent hot-loops in error situations ([#25748](https://github.com/kubernetes/kubernetes/pull/25748), [@derekwaynecarr](https://github.com/derekwaynecarr))

* Bug fixes

  * Changed Kubelet default image-gc-high-threshold to 85% to resolve a conflict with default settings in docker that prevented image garbage collection from resolving low disk space situations when using devicemapper storage. ([#40432](https://github.com/kubernetes/kubernetes/pull/40432), [@sjenning](https://github.com/sjenning))

   ([#122530](https://github.com/kubernetes/kubernetes/pull/122530), [@neolit123](https://github.com/neolit123))...

     * this is no worse than what FutureTask does in that situation. Additionally, because the
     * Future was cancelled, its listeners have been run, so its consumers will not hang.
     *
     * Contrast this to the situation we have if setResult() throws, a situation described below.
     */
    I sourceResult;
    try {
      sourceResult = getDone(localInputFuture);

 *
 * We use this because eager classpath checks cause confusion and excessive logging in Android,
 * and we can't rely on classnames after proguard, so are probably best served by falling through
 * to a situation of trying our least likely noisiest options.
 */
class DeferredSocketAdapter(
  private val socketAdapterFactory: Factory,
) : SocketAdapter {
  private var delegate: SocketAdapter? = null

     * this is no worse than what FutureTask does in that situation. Additionally, because the
     * Future was cancelled, its listeners have been run, so its consumers will not hang.
     *
     * Contrast this to the situation we have if setResult() throws, a situation described below.
     */
    I sourceResult;
    try {
      sourceResult = getDone(localInputFuture);