* needed.
   *
   * A single spinlock ("busy") is used for initializing and
   * resizing the table, as well as populating slots with new Cells.
   * There is no need for a blocking lock; when the lock is not
   * available, threads try other slots (or the base).  During these
   * retries, there is increased contention and reduced locality,
   * which is still better than alternatives.
   *

      return result;
    }

    /**
     * Following the contract of {@link AbstractQueuedSynchronizer} we create a private subclass to
     * hold the synchronizer. This synchronizer is used to implement the blocking and waiting calls
     * as well as to handle state changes in a thread-safe manner. The current state of the future
     * is held in the Sync state, and the lock is released whenever the state changes to {@link

  public final double getRate() {
    synchronized (mutex()) {
      return doGetRate();
    }
  }

  abstract double doGetRate();

  /**
   * Acquires a single permit from this {@code RateLimiter}, blocking until the request can be
   * granted. Tells the amount of time slept, if any.
   *
   * <p>This method is equivalent to {@code acquire(1)}.
   *

 * arbitrary Charset.
 *
 * <p>This is an alternative to copying the data to an {@code OutputStream} via a {@code Writer},
 * which is necessarily blocking. By implementing an {@code InputStream} it allows consumers to
 * "pull" as much data as they can handle, which is more convenient when dealing with flow
 * controlled, async APIs.
 *
 * @author Chris Nokleberg
 */

    /**
     * Records cache misses. This should be called when a cache request returns a value that was not
     * found in the cache. This method should be called by the loading thread, as well as by threads
     * blocking on the load. Multiple concurrent calls to {@link Cache} lookup methods with the same
     * key on an absent value should result in a single call to either {@code recordLoadSuccess} or

 * arbitrary Charset.
 *
 * <p>This is an alternative to copying the data to an {@code OutputStream} via a {@code Writer},
 * which is necessarily blocking. By implementing an {@code InputStream} it allows consumers to
 * "pull" as much data as they can handle, which is more convenient when dealing with flow
 * controlled, async APIs.
 *
 * @author Chris Nokleberg
 */

Only once those two conditions are met does the CNI plugin return success - effectively allowing our plugin to block Pod scheduling until we are ready.
Otherwise, the CNI plugin will continually be retried, blocking Pod startup until it succeeds.

From the above we are ensured that when the application starts and opens a connection, Ztunnel will be ready to `accept()` it.

    val pooled1 = planReusePooledConnection()
    if (pooled1 != null) return pooled1

    // Attempt a deferred plan before new routes.
    if (deferredPlans.isNotEmpty()) return deferredPlans.removeFirst()

    // Do blocking calls to plan a route for a new connection.
    val connect = planConnect()

    // Now that we have a set of IP addresses, make another attempt at getting a connection from

  public final double getRate() {
    synchronized (mutex()) {
      return doGetRate();
    }
  }

  abstract double doGetRate();

  /**
   * Acquires a single permit from this {@code RateLimiter}, blocking until the request can be
   * granted. Tells the amount of time slept, if any.
   *
   * <p>This method is equivalent to {@code acquire(1)}.
   *

type DeadlineWriter struct {
	io.WriteCloser
	timeout time.Duration
	err     error
}

// NewDeadlineWriter wraps a writer to make it respect given deadline
// value per Write(). If there is a blocking write, the returned Writer
// will return whenever the timer hits (the return values are n=0
// and err=context.DeadlineExceeded.)
func NewDeadlineWriter(w io.WriteCloser, timeout time.Duration) io.WriteCloser {