nvflare.app_opt.flower.executor module

class FlowerExecutor(start_task_name='start', configure_task_name='config', per_msg_timeout=10.0, tx_timeout=100.0, client_shutdown_timeout=5.0, extra_env: dict | None = None, allow_runtime_dependency_installation: bool = False)[source]

Bases: TieExecutor

FlowerExecutor constructor

Parameters:

  • start_task_name – name of the “start” task

  • configure_task_name – name of the “config” task

  • per_msg_timeout – per-msg timeout for ReliableMessage

  • tx_timeout – transaction timeout for ReliableMessage

  • client_shutdown_timeout – how long to wait for graceful shutdown of the client

  • extra_env – extra env variables to be passed to client applet

  • allow_runtime_dependency_installation – whether to allow dynamic dependency installation (only flwr>=1.29)

configure(config: dict, fl_ctx: FLContext)[source]

Called by the TieExecutor to configure the executor based on the config params received from the server. A subclass of TieExecutor should implement this method.

Parameters:

  • config – the config data

  • fl_ctx – FL context

Returns: None

get_applet(fl_ctx: FLContext)[source]

Called by the TieExecutor to get the Applet to be used by this executor. A subclass of TieExecutor must implement this method.

Parameters:

fl_ctx – the FL context

Returns: an Applet object

get_connector(fl_ctx: FLContext)[source]

Called by the TieExecutor to get the Connector to be used by this executor. A subclass of TieExecutor must implement this method.

Parameters:

fl_ctx – the FL context

Returns: a Connector object

get_connector_config(fl_ctx: FLContext) → dict[source]

Called by the TieExecutor to get config params for the connector. A subclass of TieExecutor should implement this method. Note that this method is always called after the “configure” method, hence it’s possible to dynamically determine the connector’s config based on the config params in the “configure” step.

Parameters:

fl_ctx – the FL context

Returns: a dict of config params