celery.md

Celery

Celery is a distributed task queue used to run background tasks.

It is optional, but is required for the "background task" example to work.

If you're using Docker in development then Celery should automatically be configured and running. The following instructions are for running Celery outside of Docker, or in production.

Quick Start

The easiest way to get going is to download and install Redis (if you don't already have it) and then run:

celery -A {{ project_name }} worker -l info --pool=solo

Note that the 'solo' pool is recommended for development but not for production. When running in production, you should use a more robust pool implementation such as prefork (for CPU bound tasks) or gevent (for I/O bound tasks).

Running Celery with Gevent

The gevent pool is useful when running tasks that are I/O bound which tends to be 90% of tasks. The same configuration can also be used to run Celery on Windows (if the solo pool is not suitable) since Celery 4.x no longer officially supports Windows.

To use the gevent pool, change the concurrency pool implementation to gevent instead.

pip install gevent
celery -A {{ project_name }} worker -l info -P gevent

For more information see the Celery documentation.

Setup and Configuration

The above setup uses Redis as a message broker and result backend. If you want to use a different message broker, for example RabbitMQ, you will need to modify the CELERY_BROKER_URL and CELERY_RESULT_BACKEND values in settings.py.

More details can be found in the Celery documentation.

Monitoring with Flower

Flower is an open-source web application for monitoring and managing Celery clusters. It provides real-time information about the status of Celery workers and tasks.

If you'd like to use Flower in development, add the following to the services section of your docker-compose.yml:

  flower:
    image: mher/flower
    environment:
      - CELERY_BROKER_URL=redis://redis:6379
    command: celery flower
    ports:
      - 5555:5555
    depends_on:
      - redis

In production you will likely want to run Flower behind a private VPN, or set up authentication on your Flower instance, and use a reverse proxy to expose it.