How can I set up Celery to call a custom initialization function before running my tasks?

Question

I have a Django project and I'm trying to use Celery to submit tasks for background processing ( http://ask.github.com/celery/introduction.html ). Celery integrates well with Django and I've been able to submit my custom tasks and get back results.

The only problem is that I can't find a sane way of performing custom initialization in the daemon process. I need to call an expensive function that loads a lot of memory before I start processing the tasks, and I can't afford to call that function every time.

Has anyone had this problem before? Any ideas how to work around it without modifying the Celery source code?

Thanks

Answer 1

You can either write a custom loader, or use the signals.

Loaders have the on_task_init method, which is called when a task is about to be executed, and on_worker_init which is called by the celery+celerybeat main process.

Using signals is probably the easiest, the signals available are:

0.8.x:

• task_prerun(task_id, task, args, kwargs)

  Dispatched when a task is about to be executed by the worker (or locally if using apply/or if CELERY_ALWAYS_EAGER has been set).

• task_postrun(task_id, task, args, kwargs, retval) Dispatched after a task has been executed in the same conditions as above.

• task_sent(task_id, task, args, kwargs, eta, taskset)

  Called when a task is applied (not good for long-running operations)

Additional signals available in 0.9.x (current master branch on github):

• worker_init()

  Called when celeryd has started (before the task is initialized, so if on a system supporting fork, any memory changes would be copied to the child worker processes).

• worker_ready()

  Called when celeryd is able to receive tasks.

• worker_shutdown()

  Called when celeryd is shutting down.

Here's an example precalculating something the first time a task is run in the process:

from celery.task import Task
from celery.registry import tasks
from celery.signals import task_prerun

_precalc_table = {}

class PowersOfTwo(Task):

    def run(self, x):
        if x in _precalc_table:
            return _precalc_table[x]
        else:
            return x ** 2
tasks.register(PowersOfTwo)


def _precalc_numbers(**kwargs):
    if not _precalc_table: # it's empty, so haven't been generated yet
        for i in range(1024):
            _precalc_table[i] = i ** 2


# need to use registered instance for sender argument.
task_prerun.connect(_precalc_numbers, sender=tasks[PowerOfTwo.name])

If you want the function to be run for all tasks, just skip the sender argument.