Control Plane

This module contains optional helpers that add extra debug and control handlers into your application.

torch.distributed.elastic.control_plane.worker_main()

This is a context manager that wraps your main entry function. This combines the existing errors.record logic as well as a new _WorkerServer that exposes handlers via a unix socket specified by Torch_WORKER_SERVER_SOCKET.

Example

@worker_main()
def main():
    pass

if __name__=="__main__":
   main()

Return type

Generator[None, None, None]