django-lightweight-queue 4.1.1

Lightweight & modular queue and cron system for Django

Django Lightweight Queue

DLQ is a lightweight & modular queue and cron system for Django. It powers millions of production jobs every day at Thread.

Installation

pip install django-lightweight-queue[redis]

Currently the only production-ready backends are redis-based, so the redis extra is essentially required. Additional non-redis backed production-ready backends are great candidates for community contributions.

Basic Usage

import time
from django_lightweight_queue import task

# Define a task
@task()
def long_running_task(first_arg, second_arg):
    time.sleep(first_arg * second_arg)

# Request that the task be executed at some point
long_running_task(4, second_arg=9)

See the docstring on the task decorator for more details.

Configuration

All automatically picked up configuration options begin with LIGHTWEIGHT_QUEUE_ and can be found in app_settings.py. They should be placed in the usual Django settings files, for example:

LIGHTWEIGHT_QUEUE_BACKEND = 'django_lightweight_queue.backends.redis.RedisBackend'

Special Configuration

If desired, specific configuration overrides can be placed in a standalone python file which passed on the command line. This is useful for applying customisations for specific servers.

For example, given a special.py containing:

LIGHTWEIGHT_QUEUE_REDIS_PORT = 12345

and then running:

$ python manage.py queue_runner --config=special.py

will result in the runner to use the settings from the specified configuration file in preference to settings from the Django environment. Any settings not present in the specified file are inherited from the global configuration.

Backends

There are four built-in backends:

Backend	Type	Description
Synchronous	Development	Executes the task inline, without any actual queuing.
Redis	Production	Executes tasks at-most-once using Redis for storage of the enqueued tasks.
Reliable Redis	Production	Executes tasks at-least-once using Redis for storage of the enqueued tasks (subject to Redis consistency). Does not guarantee the task completes.
Debug Web	Debugging	Instead of running jobs it prints the url to a view that can be used to run a task in a transaction which will be rolled back. This is useful for debugging and optimising tasks.

Running Workers

The queue runner is implemented as a Django management command:

$ python manage.py queue_runner

Workers can be distributed over multiple hosts by telling each runner that it is part of a pool:

$ python manage.py queue_runner --machine 2 --of 4

Alternatively a runner can be told explicitly which configuration to use:

$ python manage.py queue_runner --exact-configuration --config=special.py

When using --exact-configuration the number of workers is configured exactly, rather than being treated as the configuration for a pool. Additionally, exactly-configured runners will not run any cron workers.

Example

Given a Django configuration containing:

LIGHTWEIGHT_QUEUE_WORKERS = {
    'queue1': 3,
}

and a special.py containing:

LIGHTWEIGHT_QUEUE_WORKERS = {
    'queue1': 2,
}

Running any of:

$ python manage.py queue_runner --machine 1 --of 3 # or,
$ python manage.py queue_runner --machine 2 --of 3 # or,
$ python manage.py queue_runner --machine 3 --of 3

will result in one worker for queue1 on the current machine, while:

$ python manage.py queue_runner --exact-configuration --config=special.py

will result in two workers on the current machine.

Cron Tasks

DLQ supports the use of a cron-like specification of Django management commands to be run at certain times.

To specify that a management command should be run at a given time, place a cron.py file in the root folder of the Django app which defines the command and which contains a CONFIG variable:

CONFIG = (
    {
        'command': 'my_cron_command',
        # Day values 1-7 to match datetime.datetime.utcnow().isoweekday()
        'days': '*',
        'hours': '*',
        'minutes': '*',
        # Equivalent behaviour to the kwarg to `task` of the same name
        'sigkill_on_stop': True,
    },
)

Maintainers

This repository was created by Chris Lamb at Thread, and continues to be maintained by the Thread engineering team.