nvflare.apis.server_engine_spec module

class ServerEngineSpec

Bases: EngineSpec, ABC

abstract cancel_client_resources(resource_check_results: Dict[str, Tuple[bool, str]], resource_reqs: Dict[str, dict], fl_ctx: FLContext)

Cancels the request resources for the job.

Parameters:

• resource_check_results – A dict of {client_name: client_check_result} where client_check_result is a tuple of (is_resource_enough, resource reserve token if any)

• resource_reqs – A dict of {client_name: resource requirements dict}

• fl_ctx – FLContext

abstract check_client_resources(job: Job, resource_reqs: Dict[str, dict], fl_ctx: FLContext) → Dict[str, Tuple[bool, str | None]]

Sends the check_client_resources requests to the clients.

Parameters:

• job – job object

• resource_reqs – A dict of {client_name: resource requirements dict}

• fl_ctx – FLContext

Returns:

client_check_result}.

client_check_result is a tuple of (is_resource_enough, token); is_resource_enough is a bool indicates whether there is enough resources; token is for resource reservation / cancellation for this check request.

Return type:

A dict of {client_name

fire_and_forget_aux_request(targets: [], topic: str, request: Shareable, fl_ctx: FLContext, optional=False, secure=False) → dict

abstract fire_event(event_type: str, fl_ctx: FLContext)

abstract get_client_name_from_token(token: str) → str

Gets the client name from client login token.

Parameters:

token – client login token

Returns:

Client name

abstract get_clients() → List[Client]

abstract get_component(component_id: str) → object

abstract get_widget(widget_id: str) → Widget

Get the widget with the specified ID.

Parameters:

widget_id – ID of the widget

Returns: the widget or None if not found

abstract get_workspace() → Workspace

abstract new_context() → FLContext

abstract persist_components(fl_ctx: FLContext, completed: bool)

To persist the FL running components

Parameters:

• fl_ctx – FLContext

• completed – flag to indicate where the run is complete

Returns:

abstract register_aux_message_handler(topic: str, message_handle_func)

Register aux message handling function with specified topics.

Exception is raised when:

a handler is already registered for the topic; bad topic - must be a non-empty string bad message_handle_func - must be callable

Implementation Note:

This method should simply call the ServerAuxRunner’s register_aux_message_handler method.

Parameters:

• topic – the topic to be handled by the func

• message_handle_func – the func to handle the message. Must follow aux_message_handle_func_signature.

abstract restore_components(snapshot: RunSnapshot, fl_ctx: FLContext)

To restore the FL components from the saved snapshot

Parameters:

• snapshot – RunSnapshot

• fl_ctx – FLContext

Returns:

abstract send_aux_request(targets: [], topic: str, request: Shareable, timeout: float, fl_ctx: FLContext, optional=False, secure=False) → dict

Send a request to specified clients via the aux channel.

Implementation: simply calls the ServerAuxRunner’s send_aux_request method.

Parameters:

• targets – target clients. None or empty list means all clients

• topic – topic of the request

• request – request to be sent

• timeout – number of secs to wait for replies. 0 means fire-and-forget.

• fl_ctx – FL context

• optional – whether this message is optional

• secure – send the aux request in a secure way

Returns: a dict of replies (client name => reply Shareable)

abstract start_client_job(job_id, client_sites, fl_ctx: FLContext)

To send the start client run commands to the clients

Parameters:

• client_sites – client sites

• job_id – job_id

• fl_ctx – FLContext

Returns:

abstract sync_clients_from_main_process()

To fetch the participating clients from the main parent process

Returns: clients

abstract update_job_run_status()

To update the job run status to parent process.