README.md

Django celery token bucket

A dynamic token bucket implementation using the database scheduler django celery beat.

How it works

The bucket is represented by a celery queue that will not be processed by a worker but just hold our tokens (messages). Whenever a rate limited task should be run, the decorator tries to consume a message from that queue. If the queue is empty, the task gets retried after the defined timeout.
A periodic task will then refill the bucket with tokens whenever they should be available again.

Define a token bucket

Buckets are defined in the Django config.

Following example allows one thousand tokens per hour to throttle access to a rate limited third party API.

Add to settings.py of your project.

from typing import Dict
from celery import schedules
from django_celery_token_bucket.bucket import TokenBucket

INSTALLED_APPS = [
    ...,
    'django_celery_token_bucket'
]

CELERY_TOKEN_BUCKETS: Dict[str, TokenBucket] = {
    "my_api_client": TokenBucket(
        name="my_api_client",
        schedule=schedules.crontab(minute=0),  # once every hour
        amount=1000,
        maximum=1000,
    )
}

name

The name must only consist of letters, numbers and the underscore character as it's used in the name of the celery queue. It should also be the same as the key in the CELERY_TOKEN_BUCKETS dictionary.

schedule

A celery.schedules.crontab that defines when the tokens should be refilled.

amount

The amount of tokens to add whenever the scheduled refill is run.

maximum

The maximum amount of tokens our bucket can hold.

token_refill_queue

Override this setting if you want token refill tasks for this bucket to be placed on a specific queue. This field is by default None. If no value is provided, the CELERY_DEFAULT_QUEUE setting is used or celery

Sync period tasks to refill the buckets

A management command token_bucket_register_periodic_tasks is provided that should be run during deployment of your application to sync the period tasks and make sure that buckets get properly refilled.

Use the rate_limit decorator

The decorator will make sure that the task that gets decorated will not exceed the limit of available tokens.
Decorated tasks must always be bound to allow access to the task instance.

from my_app.celery import celery_app
from django_celery_token_bucket.decorators import rate_limit


@celery_app.task(bind=True)
@rate_limit(token_bucket_name="my_api_client", countdown=300)
def my_tasK(self, *args, **kwargs):
    return

token_bucket

Name of the token bucket to consume from. Must be defined in the settings (see above) or will fail with an Exception.

countdown

Time to wait in seconds before the next try when no token is available.

affect_task_retries

Defaults to False
By default a failed token retrieval will not impact the retry count of your task. To change this behavior, set affect_task_retries to True.

@celery_app.task(bind=True, max_retries=3, countdown=60)
@rate_limit(token_bucket="my_api_client", countdown=300, affect_task_retries=True)
def my_tasK(self, *args, **kwargs):
    return

In this scenario, a failed token retrieval will increase the retry count of the task decorator. If we cannot get a token on the first try, we will start over again with the 2nd try.

Run the tests locally

A docker compose environment is provided to easily run the tests:

docker compose run --rm app test

Making a new release

bumpversion is used to manage releases.

Add your changes to the CHANGELOG, run

bumpversion <major|minor|patch>

and push (including tags).