taskferry-django

Django integration for Taskferry.

flowchart BT
    DJ["Django"]
    TPD["taskferry-django"]
    TP["taskferry"]

    DJ --> TPD --> TP

Taskferry does not invent a parallel task API — Django 6 already has one. This package supplies a backend for the standard TASKS setting, so your @task code never changes while the engine behind it becomes a configuration choice.

pip install taskferry-django taskferry-procrastinate

Your code never changes

# myapp/tasks.py
from django.tasks import task


@task
def resize_image(image_id: int) -> None: ...

# anywhere
resize_image.enqueue(42)

Settings

TASKS = {"default": {"BACKEND": "taskferry_django.TaskferryBackend"}}

TASKFERRY = {
    "backends": {
        "pg": {"factory": "procrastinate", "app": "myapp.tasks:app"},
        "heavy": {"factory": "cloudrun", "project": "p", "location": "europe-west1"},
    },
    "routes": [
        {"kind": "task", "queue": "metadata", "backend": "pg"},
        {"kind": "job", "profile": "heavy", "backend": "heavy"},
    ],
    "defaults": {"task": "pg", "job": "heavy"},
}

With no TASKFERRY setting at all, everything runs locally — a fresh project works immediately and is obviously not production.

What else you get

Enqueue after commit. The classic race — a worker reaching a row before the transaction that created it commits — with the classic fix:

from django.db import transaction
from taskferry_django import task_on_commit


def create_order(request):
    with transaction.atomic():
        order = Order.objects.create(...)
        task_on_commit("myapp.tasks:process_order", order.id)

System checks. manage.py check builds every configured backend, because "the adapter is installed" and "the adapter works with these options" are different questions:

taskferry.E002  the Taskferry backend 'heavy' cannot be built: CloudRunJobBackend
               needs both 'project' and 'location'
taskferry.W003  DEBUG is False but these Taskferry backends run in this process and
               lose pending work on restart: fast

The CLI, with settings loaded.

python manage.py taskferry doctor
python manage.py taskferry capabilities pg

Jobs and inline execution, through the same runtime:

from taskferry_django import get_runtime

get_runtime().jobs.submit("build-cog", image="gdal:latest", profile="heavy")

The dependency direction

taskferry never imports Django. The test suite for the core runs — and CI runs it — in an environment where Django is not installed. The same TASKFERRY dict works verbatim in a FastAPI service, a CLI or a library.

License

Apache-2.0.