/*
     * The initialization of seenExceptionsField has to be more complicated than we'd like. The
     * simple approach would be for each caller CAS it from null to a Set populated with its
     * exception. But there's another race: If the first thread fails with an exception and a second
     * thread immediately fails with the same exception:
     *
     * Thread1: calls setException(), which returns true, context switch before it can CAS

   * on the fields of TimeoutFuture. This field is slightly different from the fields discussed
   * there: cancel() never reads this field, only writes to it. That makes the race here completely
   * harmless, rather than just 99.99% harmless.
   */
  @LazyInit private @Nullable List<@Nullable Present<V>> values;

  @SuppressWarnings("EmptyList") // ImmutableList doesn't support nullable element types

    void setNext(@Nullable Waiter next) {
      putNext(this, next);
    }

    void unpark() {
      // This is racy with removeWaiter. The consequence of the race is that we may spuriously call
      // unpark even though the thread has already removed itself from the list. But even if we did
      // use a CAS, that race would still exist (it would just be ever so slightly smaller).
      Thread w = thread;
      if (w != null) {

    void setNext(@Nullable Waiter next) {
      putNext(this, next);
    }

    void unpark() {
      // This is racy with removeWaiter. The consequence of the race is that we may spuriously call
      // unpark even though the thread has already removed itself from the list. But even if we did
      // use a CAS, that race would still exist (it would just be ever so slightly smaller).
      Thread w = thread;
      if (w != null) {

      long nanos = expirationNanos;
      long now = System.nanoTime();
      if (nanos == 0 || now - nanos >= 0) {
        synchronized (lock) {
          if (nanos == expirationNanos) { // recheck for lost race
            T t = delegate.get();
            value = t;
            nanos = now + durationNanos;
            // In the very unlikely event that nanos is 0, set it to 1;

      long nanos = expirationNanos;
      long now = System.nanoTime();
      if (nanos == 0 || now - nanos >= 0) {
        synchronized (lock) {
          if (nanos == expirationNanos) { // recheck for lost race
            T t = delegate.get();
            value = t;
            nanos = now + durationNanos;
            // In the very unlikely event that nanos is 0, set it to 1;

- Fixed a race condition in the kube-proxy initialization that could cause UDP traffic to service VIP. ([#126532](https://github.com/kubernetes/kubernetes/pull/126532), [@wedaly](https://github.com/wedaly))

        // We reschedule ourselves with a lock held for two reasons. 1. we want to make sure that
        // cancel calls cancel on the correct future. 2. we want to make sure that the assignment
        // to currentFuture doesn't race with itself so that currentFuture is assigned in the
        // correct order.
        Throwable scheduleFailure = null;
        Cancellable toReturn;
        lock.lock();
        try {

        // We reschedule ourselves with a lock held for two reasons. 1. we want to make sure that
        // cancel calls cancel on the correct future. 2. we want to make sure that the assignment
        // to currentFuture doesn't race with itself so that currentFuture is assigned in the
        // correct order.
        Throwable scheduleFailure = null;
        Cancellable toReturn;
        lock.lock();
        try {

  // introduce a very specific kind of data-race. And given the other operations performed by these
  // methods that involve volatile read/write operations, in practice there is no issue. Also, the
  // way in such a visibility issue would surface is most likely as a failure of cancel() to
  // propagate to the input. Cancellation propagation is fundamentally racy so this is fine.
  //