A Quick Guide to Celery Task Routing

By default, Celery routes all tasks to a single queue and all workers consume from this default queue. You can change this behaviour by telling Celery which tasks to send to which queues. This is known as task routing.

This is useful if you have slow and fast tasks, and you don't want the slow tasks to interfere with the fast tasks. Or you have a gevent pool worker for IO-bound tasks and a prefork pool worker for CPU-heavy tasks.

Step 1: Configure task_routes

The first thing is to assign a queue to each task. For example to route task_1 to queue_a and task_2 to queue_b:

from celery import Celery

app = Celery(
    __name__,
    broker="redis://localhost:6379/0",
    task_routes={
        "task_1": {"queue": "queue_a"}
        "task_2": {"queue": "queue_b"}
    })

Step 2: Worker queues

The --queues command line argument makes the Celery worker process tasks from one or multiple queues and ignore everything else. For the example above, I have one worker for queue_a...

# process tasks in queue_a only
$ celery --app=worker.app worker --queues=queue_a

...and another worker processing tasks from queue_b:

# process tasks in queue_b only
$ celery --app=worker.app worker --queues=queue_b

Bonus: Multiple queues

If you want your worker to process tasks from more than one queue, pass a comma-separated list to the --queuesargument:

$ celery --app=worker.app worker --queues=queue_a,queue_b

Step 3: Give it a go

You can find the complete example code on GitHub. Clone the repository, create a virtual environment and install the dependencies:

$ git clone https://github.com/bstiel/celery-task-routing.git
$ cd celery-task-routing
$ python -m venv venv
$ source venv/bin/activate
$ pip install -r requirements.txt

Start the Redis message broker:

$ docker compose up -d

Start the two Celery workers and the producer, to produce task_1 and task_2 once a second, inside a look. All three commands are wrapped in a Procfile that you can start via honcho (or foreman):

$ honcho start

You will see worker_1 processing task_1 only, and worker_2 processing task_2 only.

11:42:16 worker_1.1  [2023-12-20 11:42:16,904: WARNING/MainProcess] hello from task_1
11:42:16 worker_2.1  [2023-12-20 11:42:16,913: WARNING/MainProcess] hello from task_2

Summary

In this blog post, you learned how to configure Celery to route tasks to dedicated queues and make Celery workers process tasks from certain queues only.

To achieve that, you need to map tasks to queues. This approach works as long as you only have a limited number of tasks. For more complex setups involving many tasks, queues and even services, dynamic task routing offers a more scalable and maintainable solution.

Last updated Dec 19, 2023
First published May 29, 2018