*       "safe" edge is created.
 *   <li>If a cycle is detected, an "unsafe" (cyclic) edge is created to represent a potential
 *       deadlock situation, and the appropriate Policy is executed.
 * </ul>
 *
 * <p>Note that detection of potential deadlock does not necessarily indicate that deadlock will
 * happen, as it is possible that higher level application logic prevents the cyclic lock

                        const int in_flight_nodes_limit)
      : status_(TF_NewStatus()),
        // If the context's default exector is set to async, re-using that in
        // each thread would cause collectives to deadlock. For consistency we
        // create a new sync executor for every thread.
        //
        // TODO(allenl): We should have an async API that works with the
        // parallel device.
        device_(device),

  // parallel device, then call `Join` on each; even if some of the `Join`s
  // return a bad status the caller must run all of the `Join`s or any future
  // `StartExecute`s will deadlock).
  //
  // If `is_async=false` (constructor argument), `cancellation_manager` must
  // live until `Join` finishes. If `is_async=true` it must live until `Join` is

  // threadpool, so that we avoid the possibility of running the runner_ in the
  // threadpool of GPU event mgr, as that can trigger more callbacks to be
  // scheduled on that same threadpool, causing a deadlock in cases where the
  // caller of event_mgr->ThenExecute() blocks on the completion of the callback
  // (as in the case of ConstOp kernel creation on GPU, which involves copying a
  // CPU tensor to GPU).

    assertSame(firstException.getCause(), expected.getCause());
    // lockA should work after lockB is released.
    lockB.unlock();
    lockA.lock();
  }

  // Tests transitive deadlock detection.
  public void testDeadlock_threeLocks() {
    // Establish an ordering from lockA -> lockB.
    lockA.lock();
    lockB.lock();
    lockB.unlock();
    lockA.unlock();

    assertSame(firstException.getCause(), expected.getCause());
    // lockA should work after lockB is released.
    lockB.unlock();
    lockA.lock();
  }

  // Tests transitive deadlock detection.
  public void testDeadlock_threeLocks() {
    // Establish an ordering from lockA -> lockB.
    lockA.lock();
    lockB.lock();
    lockB.unlock();
    lockA.unlock();

                this.state = 6;
                ioe = ioe0;
            }
        case 4: /* failed to connect - reset the transport */
            // thread is cleaned up by connect routine, joining it here causes a deadlock
            this.thread = null;
            this.state = 6;
            break;
        default:
            log.error("Invalid state: " + this.state);
            this.thread = null;

   * This covers a bug where if a listener was installed that would stop the manager if any service
   * fails and something failed during startup before service.start was called on all the services,
   * then awaitStopped would deadlock due to an IllegalStateException that was thrown when trying to
   * stop the timer(!).
   */
  public void testFailStart_stopOthers() throws TimeoutException {
    Service a = new FailStartService();

   * under {@code directExecutor}: Not only might the wait for a lock be long, but if the running
   * thread was holding a lock, the listener may deadlock or break lock isolation.
   *
   * <p>This instance is equivalent to:
   *
   * <pre>{@code
   * final class DirectExecutor implements Executor {
   *   public void execute(Runnable r) {
   *     r.run();

  /**
   * Returns the stripes that correspond to the passed objects, in ascending (as per {@link
   * #getAt(int)}) order. Thus, threads that use the stripes in the order returned by this method
   * are guaranteed to not deadlock each other.
   *
   * <p>It should be noted that using a {@code Striped<L>} with relatively few stripes, and {@code
   * bulkGet(keys)} with a relative large number of keys can cause an excessive number of shared