// custom devices will fail.
//
// `device_name` must not name an existing physical or custom device. It must
// follow the format:
//
//    /job:<name>/replica:<replica>/task:<task>/device:<type>:<device_num>
//
// If the device is successfully registered, `status` is set to TF_OK. Otherwise
// the device is not usable. In case of a bad status, `device.delete_device` is

  // on all tasks in the cluster. This call internally coordinates with other
  // tasks to initialize the eager context and TF server for multi-client
  // execution.
  virtual Status EnableCollectiveOps(const ServerDef& server_def) = 0;

  // Check if the remote task is alive.
  virtual Status CheckRemoteAlive(const std::string& remote_task_name,
                                  bool* is_alive) = 0;

tensorflow::ServerDef GetServerDef(const tensorflow::string& job_name,
                                   int num_tasks);

// Create a ServerDef with job name "localhost" and add `num_tasks` tasks in it.
tensorflow::ServerDef GetServerDef(int num_tasks);

// Create a multi-client ServerDef with the given `job_name`, add `num_tasks`

      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:

// 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;

// A Server instance encapsulates a set of devices and a Session target that
// can participate in distributed training. A server belongs to a cluster
// (specified by a ClusterSpec), and corresponds to a particular task in a
// named job. The server can communicate with any other server in the same
// cluster.

// In-process TensorFlow server.
typedef struct TF_Server TF_Server;

  // all tasks in the cluster.
  // This call internally coordinates with other tasks to initialize the eager
  // context and TF server for multi-client execution.
  virtual Status EnableCollectiveOps(const ServerDef& server_def) = 0;

  // Set a distributed manager that helps set up, update, and check liveness
  // of member tasks in the cluster.
  virtual void SetDistributedManager(