/*
       * Assume that equals uses the == optimization when appropriate, and that
       * it would check hash codes as an optimization when appropriate. If we
       * did these things, it would just make things worse for the most
       * performance-conscious users.
       */
      if (key.equals(candidateKey)) {
        return entry.getValue();
      }
    }
    return null;
  }

  @Override

 *       method to obtain a correctly-sized copy.
 *   <li>The performance of using the associated {@code Builder} class can be assumed to be no
 *       worse, and possibly better, than creating a mutable collection and copying it.
 *   <li>Implementations generally do not cache hash codes. If your element or key type has a slow

 *       method to obtain a correctly-sized copy.
 *   <li>The performance of using the associated {@code Builder} class can be assumed to be no
 *       worse, and possibly better, than creating a mutable collection and copying it.
 *   <li>Implementations generally do not cache hash codes. If your element or key type has a slow

       * (We also haven't benchmarked under Android. We continue to use UnsafeAtomicHelper there so
       * that we don't change the performance there, for better or for worse.) Fortunately, JVM
       * users will typically use guava-jre, not guava-android, and guava-jre uses the VarHandle
       * implementation when possible.
       */
      try {

 *
 * <p><i>Performance notes:</i>
 *
 * <ul>
 *   <li>If you only access one end of the queue, and do use a maximum size, this class will perform
 *       significantly worse than a {@code PriorityQueue} with manual eviction above the maximum
 *       size. In many cases {@link Ordering#leastOf} may work for your use case with significantly
 *       improved (and asymptotically superior) performance.

 *
 * <p><i>Performance notes:</i>
 *
 * <ul>
 *   <li>If you only access one end of the queue, and do use a maximum size, this class will perform
 *       significantly worse than a {@code PriorityQueue} with manual eviction above the maximum
 *       size. In many cases {@link Ordering#leastOf} may work for your use case with significantly
 *       improved (and asymptotically superior) performance.

</pre>

<p>
As before, there is no guarantee that, in <code>main</code>,
observing the write to <code>done</code>
implies observing the write to <code>a</code>, so this program could
print an empty string too.
Worse, there is no guarantee that the write to <code>done</code> will ever
be observed by <code>main</code>, since there are no synchronization
events between the two threads.  The loop in <code>main</code> is not
guaranteed to finish.

   *       threads, RPC network threads, or other latency-sensitive threads. In those cases, slow
   *       listeners can harm responsiveness, slow the system as a whole, or worse. (See also the
   *       note about locking below.)
   *   <li>If many tasks will be triggered by the same event, one heavyweight task may delay other

Multi-threaded tests are often not reproducible due to the fact that there is
no guarantee on the order in which operations occur across threads. Adding
randomization to the mix usually makes things worse and should be done with
care.

== Test coverage analysis

Generating test coverage reports for Elasticsearch is currently not possible through Gradle.

- Previously, when trying to allocate devices through DRA for a node timed out, scheduling would proceed with another node if any had the necessary resources. This potentially hid that a node was ignored. Worse, if scheduling was slow overall, the pod was incorrectly moved to "unschedulable" and only retried after a periodic sweep. Now timeouts are errors that are always visible as pod scheduling failures and get retried with per-pod exponential backoff....