// 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).
  // Setting a larger thread pool does not help with the Swift caller, as we use

bool ExtendSessionGraphHelper(TF_Session* session, TF_Status* status) {
  if (session->graph != nullptr) {
    // Take the graph lock before the session lock to avoid deadlock. This is
    // safe since session->graph does not change.
    session->graph->mu.lock();
    mutex_lock session_lock(session->mu);
    const Graph& graph = session->graph->graph;

    const string& mutation_warning = session->graph->sessions[session];

                    absl::optional<int64_t> step_id = absl::nullopt);
  // Block until the previous `StartExecute` operation has executed. Forwards
  // the status from `TFE_Execute` and returns outputs if the status is OK.
  std::vector<TensorHandlePtr> Join(TF_Status* status);

  // Block until all Ops finished running on the thread.
  void AsyncWait(TF_Status* status);

 private:
  void Run();

}

int TF_GraphNumFunctions(TF_Graph* g) {
  tensorflow::mutex_lock l(g->mu);
  return g->graph.flib_def().num_functions();
}

int TF_GraphGetFunctions(TF_Graph* g, TF_Function** funcs, int max_func,
                         TF_Status* status) {
  tensorflow::FunctionDefLibrary lib;
  {
    tensorflow::mutex_lock l(g->mu);
    lib = g->graph.flib_def().ToProto();
  }

        "Coordination service is not enabled.");
    return;
  }
  absl::Duration timeout;
  if (timeout_in_ms > 0) {
    timeout = absl::Milliseconds(timeout_in_ms);
  } else {
    // Block until the key-value is set or the worker shuts down.
    timeout = absl::InfiniteDuration();
  }
  auto status_or_value = coord_agent->GetKeyValue(key, timeout);
  status->status = status_or_value.status();

  buf->Unref();
  return tensorflow::wrap(tensorflow::TensorHandle::CreateLocalHandle(
      std::move(t), device, device, context));
}

// This function will block till the operation that produces `h` has
// completed. This is only valid on local TFE_TensorHandles. Returns the size in
// bytes of the memory pointed to by the device pointer returned above.