openfl.transport.grpc.director_client.ShardDirectorClient

class openfl.transport.grpc.director_client.ShardDirectorClient(*, director_host, director_port, shard_name, tls=True, root_certificate=None, private_key=None, certificate=None)

Bases: object

The internal director client class.

This class communicates with the director to manage the shard’s participation in the federation.

Class Attributes:

  • shard_name (str) – The name of the shard.

  • stub (director_pb2_grpc.DirectorStub) – The gRPC stub for communication with the director.

Methods

get_experiment_data

Get an experiment data from the director.

report_shard_info

Report shard info to the director.

send_health_check

Send envoy health check.

set_experiment_failed

Set the experiment failed.

wait_experiment

Wait an experiment data from the director.
get_experiment_data(experiment_name)

Get an experiment data from the director.

Parameters:

experiment_name (str) – The name of the experiment.

Returns:

data_stream (grpc._channel._MultiThreadedRendezvous) – The data stream of the experiment data.

report_shard_info(shard_descriptor: Type[ShardDescriptor], cuda_devices: tuple) → bool

Report shard info to the director.

Parameters:

  • shard_descriptor (Type[ShardDescriptor]) – The descriptor of the shard.

  • cuda_devices (tuple) – The CUDA devices available on the shard.

Returns:

acknowledgement (bool) – Whether the report was accepted by the director.

send_health_check(*, envoy_name: str, is_experiment_running: bool, cuda_devices_info: List[dict] | None = None) → int

Send envoy health check.

Parameters:

  • envoy_name (str) – The name of the envoy.

  • is_experiment_running (bool) – Whether an experiment is currently running.

  • cuda_devices_info (List[dict], optional) – Information about the CUDA devices. Defaults to None.

Returns:

health_check_period (int) – The period for health checks.

set_experiment_failed(experiment_name: str, error_code: int = 1, error_description: str = '')

Set the experiment failed.

Parameters:

  • experiment_name (str) – The name of the experiment.

  • error_code (int, optional) – The error code. Defaults to 1.

  • error_description (str, optional) – The description of the error. Defaults to ‘’.

wait_experiment()

Wait an experiment data from the director.

Returns:

experiment_name (str) – The name of the experiment.