// multi worker collective ops to detect failures in the cluster.  If a peer is
// down, collective ops may hang.
TF_CAPI_EXPORT extern void TFE_CollectiveOpsCheckPeerHealth(
    TFE_Context* ctx, const char* task, int64_t timeout_in_ms,
    TF_Status* status);

// Information about the shape of a Tensor and its type.
struct TF_ShapeAndType {
  // Number of dimensions. -1 indicates unknown rank.
  int num_dims;

    // During execution:
    //   * task:0 processes main function `VariableAddFunctionWithGraphError`
    //     and places the 'read0_maybe_with_graph_error' op on task:2
    //   * task:0 partitions the main function with a subgraph containing
    //     'read0_maybe_with_graph_error' sent to task:2
    //   * task:2 graph pass reports an error when it sees

  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

  const char* device_name = "/job:localhost/replica:0/task:0/device:CUSTOM:0";
  const char* first_device_name =
      "/job:localhost/replica:0/task:0/device:CPU:0";
  const char* second_device_name =
      "/job:localhost/replica:0/task:0/device:CPU:1";
  std::array<const char*, 2> underlying_devices{first_device_name,

      TF_Status* status) const;

  void AsyncWait(TFE_Context* context, TF_Status* status) const;

  // Device strings for component devices that only include a
  // worker/task/replica if any of those differ across components. Useful for
  // printing debug messages.
  std::vector<std::string> SummarizeDeviceNames() const;

 private:

  // Register the custom device
  std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)> status(
      TF_NewStatus(), TF_DeleteStatus);
  const char* device_name = "/job:localhost/replica:0/task:0/device:CUSTOM:0";
  std::array<const char*, 2> underlying_devices{first_device, second_device};
  RegisterParallelDevice(context, device_name, underlying_devices,
                         status.get());