Frequently Asked Questions¶

Open a new issue if you have a question not answered in the FAQ, Guides, or API docs.

Logging¶

How to enable agent logging in the Manager?¶

The Manager does not configure logging when an agent starts on a worker within an executor. We recommend using the worker initialization features of executors to configure logging, such as by calling init_logging() or logging.basicConfig(). For example, use the initializer argument when using a ProcessPoolExecutor.

import logging
import multiprocessing
from concurrent.futures import ProcessPoolExecutor
from academy.logging import init_logging
from academy.manager import Manager

mp_context = multiprocessing.get_context('spawn')
executor = ProcessPoolExecutor(
    max_workers=3,
    initializer=init_logging,
    initargs=(logging.INFO,),
    mp_context=mp_context,
)

async with await Manager(..., executors=executor) as manager:
    ...

Best Practices¶

This section highlights common best practices for developing applications in Academy.

Avoid communication operations during agent initialization¶

The __init__ method of an Agent is called in one of two places:

On the client when submitting an agent to be executed.
On the worker when agent instantiation is deferred.

In both scenarios, it is unsafe to perform communication operations (i.e., invoking an action on a remote agent) in __init__ because connection resources and background tasks have not yet been initialized.

The Agent.agent_on_startup() callback can be used instead to perform communication once the agent is in a running state.