reference/generated/torchrl.modules.inference_server.MPTransport

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

MPTransport

class torchrl.modules.inference_server.MPTransport(ctx: BaseContext | None = None, *, use_manager: bool = False)

Cross-process transport using multiprocessing queues.

Response routing uses per-actor queues (one per client() call) so that no mp.Queue object is ever serialised through another queue. Clients must be created with client() before spawning child processes.

Parameters:

• ctx – a multiprocessing context (e.g. mp.get_context("spawn")). Defaults to mp.get_context("spawn").

• use_manager (bool, optional) – if True, back the request and response queues with a multiprocessing manager. This is useful when clients are forwarded through another spawned process. Defaults to False.

Example

>>> import multiprocessing as mp
>>> transport = MPTransport()
>>> client = transport.client()           # creates response queue
>>> p = mp.Process(target=actor_fn, args=(client,))
>>> p.start()                             # queue inherited

client() → MailboxClient

Create an actor-side client with a dedicated response queue.

Must be called in the parent process before spawning children. In particular, a manager-backed transport (use_manager=True) loses its manager handle when pickled, so calling this on an unpickled copy raises a RuntimeError.

Returns:

A _QueueInferenceClient that can be passed to a child process as an argument to multiprocessing.Process.

close() → None

Release transport resources.

Shuts down the multiprocessing manager backing the request/response queues when the transport was built with use_manager=True (a no-op otherwise). The process that owns the transport must call this once the server and all clients are done with it: ProcessInferenceServer does not close the transport on shutdown().