Django Tasks

An implementation and backport of background workers and tasks in Django, as defined in DEP 0014.

Warning: This package is under active development, and breaking changes may be released at any time. Be sure to pin to specific versions if you're using this package in a production environment.

Installation

python -m pip install django-tasks

The first step is to add django_tasks to your INSTALLED_APPS.

INSTALLED_APPS = [
    # ...
    "django_tasks",
]

Secondly, you'll need to configure a backend. This connects the tasks to whatever is going to execute them.

If omitted, the following configuration is used:

TASKS = {
    "default": {
        "BACKEND": "django_tasks.backends.immediate.ImmediateBackend"
    }
}

A few backends are included by default:

django_tasks.backends.dummy.DummyBackend: Don't execute the tasks, just store them. This is especially useful for testing.
django_tasks.backends.immediate.ImmediateBackend: Execute the task immediately in the current thread
django_tasks.backends.database.DatabaseBackend: Store tasks in the database (via Django's ORM), and retrieve and execute them using the db_worker management command

Note: DatabaseBackend additionally requires django_tasks.backends.database adding to INSTALLED_APPS.

Usage

Note: This documentation is still work-in-progress. Further details can also be found on the DEP. The tests are also a good exhaustive reference.

Defining tasks

A task is created with the task decorator.

from django_tasks import task


@task()
def calculate_meaning_of_life() -> int:
    return 42

The task decorator accepts a few arguments to customize the task:

priority: The priority of the task (between -100 and 100. Larger numbers are higher priority. 0 by default)
queue_name: Whether to run the task on a specific queue
backend: Name of the backend for this task to use (as defined in TASKS)
enqueue_on_commit: Whether the task is enqueued when the current transaction commits successfully, or enqueued immediately. By default, this is handled by the backend (see below). enqueue_on_commit may not be modified with .using.

These attributes (besides enqueue_on_commit) can also be modified at run-time with .using:

modified_task = calculate_meaning_of_life.using(priority=10)

In addition to the above attributes, run_after can be passed to specify a specific time the task should run. Both a timezone-aware datetime or timedelta may be passed.

Enqueueing tasks

To execute a task, call the enqueue method on it:

result = calculate_meaning_of_life.enqueue()

The returned TaskResult can be interrogated to query the current state of the running task, as well as its return value.

If the task takes arguments, these can be passed as-is to enqueue.

Transactions

By default, tasks are enqueued after the current transaction (if there is one) commits successfully (using Django's transaction.on_commit method), rather than enqueueing immediately.

This can be configured using the ENQUEUE_ON_COMMIT setting. True and False force the behaviour.

TASKS = {
    "default": {
        "BACKEND": "django_tasks.backends.immediate.ImmediateBackend",
        "ENQUEUE_ON_COMMIT": False
    }
}

This can also be configured per-task by passing enqueue_on_commit to the task decorator.

Queue names

By default, tasks are enqueued onto the "default" queue. When using multiple queues, it can be useful to constrain the allowed names, so tasks aren't missed.

TASKS = {
    "default": {
        "BACKEND": "django_tasks.backends.immediate.ImmediateBackend",
        "QUEUES": ["default", "special"]
    }
}

Enqueueing tasks to an unknown queue name raises InvalidTaskError.

To disable queue name validation, set QUEUES to [].

The database backend worker

First, you'll need to add django_tasks.backends.database to INSTALLED_APPS:

INSTALLED_APPS = [
    # ...
    "django_tasks",
    "django_tasks.backends.database",
]

Then, run migrations:

./manage.py migrate

Next, configure the database backend:

TASKS = {
    "default": {
        "BACKEND": "django_tasks.backends.database.DatabaseBackend"
    }
}

Finally, you can run the db_worker command to run tasks as they're created. Check the --help for more options.

./manage.py db_worker

Pruning old tasks

After a while, tasks may start to build up in your database. This can be managed using the prune_db_task_results management command, which deletes completed tasks according to the given retention policy. Check the --help for the available options.

Retrieving task result

When enqueueing a task, you get a TaskResult, however it may be useful to retrieve said result from somewhere else (another request, another task etc). This can be done with get_result (or aget_result):

result_id = result.id

# Later, somewhere else...
calculate_meaning_of_life.get_result(result_id)

A result id should be considered an opaque string, whose length could be up to 64 characters. ID generation is backend-specific.

Only tasks of the same type can be retrieved this way. To retrieve the result of any task, you can call get_result on the backend:

from django_tasks import default_task_backend

default_task_backend.get_result(result_id)

Return values

If your task returns something, it can be retrieved from the .return_value attribute on a TaskResult. Accessing this property on an unfinished task (ie not SUCCEEDED or FAILED) will raise a ValueError.

assert result.status == ResultStatus.SUCCEEDED
assert result.return_value == 42

If a result has been updated in the background, you can call refresh on it to update its values. Results obtained using get_result will always be up-to-date.

assert result.status == ResultStatus.NEW
result.refresh()
assert result.status == ResultStatus.SUCCEEDED

Exceptions

If a task raised an exception, its .exception will be the exception raised:

assert isinstance(result.exception, ValueError)

As part of the serialization process for exceptions, some information is lost. The traceback information is reduced to a string that you can print to help debugging:

assert isinstance(result.traceback, str)

The stack frames, globals() and locals() are not available.

If the exception could not be serialized, the .result is None.

Backend introspecting

Because django-tasks enables support for multiple different backends, those backends may not support all features, and it can be useful to determine this at runtime to ensure the chosen task queue meets the requirements, or to gracefully degrade functionality if it doesn't.

supports_defer: Can tasks be enqueued with the run_after attribute?
supports_async_task: Can coroutines be enqueued?
supports_get_result: Can results be retrieved after the fact (from any thread / process)?

from django_tasks import default_task_backend

assert default_task_backend.supports_get_result

This is particularly useful in combination with Django's system check framework.

Signals

A few Signals are provided to more easily respond to certain task events.

Whilst signals are available, they may not be the most maintainable approach.

django_tasks.signals.task_enqueued: Called when a task is enqueued. The sender is the backend class. Also called with the enqueued task_result.
django_tasks.signals.task_finished: Called when a task finishes (SUCCEEDED or FAILED). The sender is the backend class. Also called with the finished task_result.

Contributing

See CONTRIBUTING.md for information on how to contribute.