* recommended that users of this method override {@link CacheLoader#reload} with an asynchronous
   * implementation; otherwise refreshes will be performed during unrelated cache read and write
   * operations.
   *
   * <p>Currently automatic refreshes are performed when the first stale request for an entry
   * occurs. The request triggering refresh will make a synchronous call to {@link

    if (forward_input) {
      // Since the input was forwarded, we released the input handle right away
      // and hence expect the input to be forwarded to the return tensor.
      EXPECT_EQ(orig_ptr, TF_TensorData(t));
    } else {
      // In async mode we expect forwarding to work without releasing the input
      // handle since by the time the kernel is executed we have released the
      // handle in the client code.

   * recommended that users of this method override {@link CacheLoader#reload} with an asynchronous
   * implementation; otherwise refreshes will be performed during unrelated cache read and write
   * operations.
   *
   * <p>Currently automatic refreshes are performed when the first stale request for an entry
   * occurs. The request triggering refresh will make a synchronous call to {@link

     */
    @Override
    protected void finalize() throws Throwable {
        try {
            if (isConnected() && this.usageCount.get() != 0) {
                log.warn("Session was not properly released, performing emergency cleanup: " + this);
                // Perform emergency cleanup to prevent memory leaks
                emergencyCleanup();
            }
        } catch (Exception e) {

 *     .build();
 * Response response = eagerClient.newCall(request).execute();
 * ```
 *
 * ## Shutdown Isn't Necessary
 *
 * The threads and connections that are held will be released automatically if they remain idle. But
 * if you are writing a application that needs to aggressively release unused resources you may do
 * so.
 *

 *  Fix: Don't release HTTP/2 connections that have multiple canceled calls. We had a bug where
    canceling calls would cause the shared HTTP/2 connection to be unnecessarily released. This
    harmed connection reuse.
 *  Fix: Ensure canceled and discarded HTTP/2 data is not permanently counted against the limited
    flow control window. We had a few bugs where window size accounting was broken when streams