Linux will then send new connections to one of the processes (first to accept wins, so effectively random).

This still requires delicate sequencing.
Consider a rolling restart of the DaemonSet; the default configuration for Istio's installation.
We will do the following:

1. `ztunnel-new` starts, connects to CNI.

  @GwtIncompatible // BlockingQueue
  public static <E> int drain(
      BlockingQueue<E> q, Collection<? super E> buffer, int numElements, Duration timeout)
      throws InterruptedException {
    // TODO(b/126049426): Consider using saturateToNanos(timeout) instead.
    return drain(q, buffer, numElements, timeout.toNanos(), NANOSECONDS);
  }

  /**

 * ListenableFuture} directly.
 *
 * <p>For interoperability between {@code ListenableFuture} and <b>{@code CompletableFuture}</b>,
 * consider <a href="https://github.com/lukas-krecan/future-converter">Future Converter</a>.
 *
 * @author Sven Mawson
 * @since 10.0 (replacing {@code Futures.makeListenable}, which existed in 1.0)
 */
@J2ktIncompatible

            /*
             * StandardDescendingMap uses lowerEntry(), and TreeMap.lowerEntry() deliberately
             * produces immutable entries.
             *
             * TODO(cpovirk): Consider making StandardDescendingMap return a ForwardingEntry that
             * supports setValue().
             */
            .suppressing(
                MapEntrySetTester.getSetValueMethod(),

     * for the next update or finish.
     *
     * @param    input    input block
     * @param    offset    start of meaningful bytes in input
     * @param    len        count of bytes in input block to consider
     */
    public void engineUpdate (byte[] input, int offset, int len) {
        // make sure we don't exceed input's allocated size/length
        if (offset < 0 || len < 0 || (long)offset + len > input.length)

  public static <E> int drain(
      BlockingQueue<E> q, Collection<? super E> buffer, int numElements, Duration timeout)
      throws InterruptedException {
    // TODO(b/126049426): Consider using saturateToNanos(timeout) instead.
    return drain(q, buffer, numElements, timeout.toNanos(), NANOSECONDS);
  }

  /**

   * @param executor the executor in which the listeners callback methods will be run. For fast,
   *     lightweight listeners that would be safe to execute in any thread, consider {@link
   *     MoreExecutors#directExecutor}.
   * @since 13.0
   */
  void addListener(Listener listener, Executor executor);

  /**
   * The lifecycle states of a service.
   *

      <T> Facade<T> newFacade() {
        return new OldAbstractFutureFacade<T>();
      }
    };

    abstract <T> Facade<T> newFacade();
  }

  @SuppressWarnings("ThreadPriorityCheck") // TODO: b/175898629 - Consider onSpinWait.
  static void awaitWaiting(Thread t) {
    while (true) {
      Thread.State state = t.getState();
      switch (state) {
        case RUNNABLE:
        case BLOCKED:
          Thread.yield();

    result.timer = scheduledExecutor.schedule(fire, time, unit);
    delegate.addListener(fire, directExecutor());
    return result;
  }

  /*
   * Memory visibility of these fields. There are two cases to consider.
   *
   * 1. visibility of the writes to these fields to Fire.run:
   *
   * The initial write to delegateRef is made definitely visible via the semantics of

I'll show you some concrete examples in the next chapters.

---

Then the next concepts to consider are all about the program running your actual API (e.g. Uvicorn).

## Program and Process