SQLAlchemy session handling in delayed Celery tasks

问题

I am using a relational database via SQLAlchemy. I want to spawn a job that deals with databases using Celery. There is a code:

from sqlalchemy.orm.session import Session
from celery.task import task
from myapp.user import User

@task
def job(user):
    # job...
    session = Session.object_session(user)
    with user.begin():
        user.value = result_value

def ordinary_web_request_handler(uid):
    assert isinstance(session, Session)
    user = session.query(User).get(int(uid))
    # deals with user...
    job.delay(user)
    return response

I need to use a SQLAlchemy session in the delayed job, but there’s no session yet. How can I set a session into passed user entity?

There are ways I thought, but I am not sure which of these (or none of these) is the best practice:

• Always pass only primary keys and retrieve new instances by the passed primary keys.
• Set the session of the passed instances (but I don’t know how).
• Do not use ORM in the delayed task.

回答1:

You should be able to initiate your session with the worker signals: http://celery.readthedocs.org/en/latest/userguide/signals.html#worker-signals

If you use a singleton-like pattern to make sure you always have a session available in that thread than everything should work just fine.

来源：https://stackoverflow.com/questions/8034098/sqlalchemy-session-handling-in-delayed-celery-tasks