Django transaction support for Celery tasks.

pip install django-celery-transactions==0.3.6



Travis Version Downloads Coverage Status

django-celery-transactions holds on to Celery tasks until the current database transaction is committed, avoiding potential race conditions as described in Celery's user guide. Send tasks from signal handlers without fear!

Django and Celery coverage

  • Django 1.6, 1.7 and 1.8 are supported and tested. 1.9 will likely require refactoring.
  • Celery 3.0.x and 3.1.x are supported, but the tests are running with 3.1.18


  • If the transaction is rolled back, the tasks are discarded. Django's transaction middleware does this if an exception is raised.
  • If transactions aren't being managed, tasks are sent as normal. This means that sending tasks from within Django's shell will work as expected, as will the various transaction decorators commit_manually, commit_on_success, etc.

Installation & Use

  1. Install django-celery-transactions from PyPI:
$ pip install django-celery-transactions
  1. Use the patched decorator to create your tasks:
from djcelery_transactions import task
from models import Model

def print_model(model_pk):
    print Model.objects.get(pk=model_pk)
  1. Then use them as normal:
from django.db import transaction
from models import Model
from tasks import print_model

# This task will be sent after the transaction is committed. This works
# from anywhere in the managed transaction block (e.g. signal handlers).
def view(request):
    model = Model.objects.create(...)

# This task will not be sent (it is discarded) as the transaction
# middleware rolls the transaction back when it catches the exception.
def failing_view(request, model_pk):
    raise Exception()

# This task will be sent immediately as transactions aren't being
# managed and it is assumed that the author knows what they're doing.
def manual_view(request, model_pk):


Due to the task being sent after the current transaction has been commited, the PostTransactionTask provided in this package does not return an celery.result.AsyncResult as the original celery Task does.

Thus, print_model.delay(model_pk) simply returns None. In order to track the task later on, the task_id can be predefined in the apply_async method:

    from celery.utils import uuid

    u = uuid()
    print_model.apply_async((model_pk), {}, task_id=u)

Compatibility with CELERY_ALWAYS_EAGER

There are 2 main reasons for CELERY_ALWAYS_EAGER:

  1. Running task synchronously and returning EagerResult, Celery's user guide

  2. Being able to run code (often tests) without a celery broker.

For this second reason, the intended behavior will often conflict with transactions handling, which is why you should then also use CELERY_EAGER_TRANSACTION


Run test suite

$ python test