Skip to main content

Django app for managing external side effects.

Project description

Django Side Effects

Django app for managing external side effects.


This project now supports Python 3.8+ and Django 3.2+ only on master.

Legacy versions are tagged.


This project was created to try and bring some order to the use of external side-effects within the YunoJuno platform. External side-effects are (as defined by us) those actions that affect external systems, and that are not part of the core application integrity. They fall into two main categories within our application - notifications and updates, and are best illustrated by example:


  • Slack messages

  • SMS (via Twilio)

  • Push notifications

  • Email


  • Base CRM (sales)

  • Mailchimp CRM (marketing)

  • Elasticsearch (full-text index)

There are some shared aspects of all of these side-effects:

  1. They can all be processed asynchronously (queued)

  2. They can all be replayed (and are idempotent)

  3. They can be executed in any order

  4. They are not time critical

  5. They do not affect the state of the Django application

As we have continued to build out YunoJuno our use of these side-effects has become ever more complex, and has in some areas left us with functions that are 80% side-effects:

def foo():
    # do the thing the function is supposed to do
    # spend the rest of the function working out which side-effects to fire
    if settings.notify_account_handler:
    if obj.has_changed_foo():

This results in a codebase is:

  • Hard to read

  • Hard to test

  • Hard to document^

^ Barely a week goes by without someone asking “what happens when X does Y - I thought they got email Z?”


This project aims to address all three of the issues above by:

  • Removing all side-effects code from core functions

  • Simplifying mocking / disabling of side-effects in tests

  • Simplifying testing of side-effects only

  • Automating documentation of side-effects

It does this with a combination of function decorators that can be used to build up a global registry of side-effects.

The first decorator, has_side_effects, is used to mark a function as one that has side effects:

# mark this function as one that has side-effects. The label
# can be anything, and is used as a dict key for looking up
# associated side-effects functions
def foo(*args, **kwargs):

Decorating view functions

By default, the has_side_effects decorator will run so long as the inner function does not raise an exception. View functions, however, are a paticular case where the function may run, and return a perfectly valid HttpResponse object, but you do not want the side effects to run, as the response object has a status_code of 404, 500, etc. In this case, you want to inspect the inner function return value before deciding whether to fire the side effects functions. In order to support this, the has_side_effects decorator has a kwarg run_on_exit which takes a function that takes a single parameter, the return value from the inner function, and must return True or False which determines whether to run the side effects.

The decorators module contains the default argument for this kwarg, a function called http_response_check. This will return False if the inner function return value is an HttpResponse object with a status code in the 4xx-5xx range.

The second decorator, is_side_effect_of, is used to bind those functions that implement the side effects to the origin function:

# bind this function to the event 'update_profile'
def send_updates(*args, **kwargs):
    """Update CRM system."""

# bind this function also to 'update_profile'
def send_notifications(*args, **kwargs):
    """Notify account managers."""

In the above example, the updates and notifications have been separated out from the origin function, which is now easier to understand as it is only responsible for its own functionality. In this example we have two side-effects bound to the same origin, however this is an implementation detail - you could have a single function implementing all the side-effects, or split them out further into the individual external systems.

Passing origin function return value to side-effects handlers

By default, side-effects handling functions must have the same function signature as the origin function. (Internally the (*args, **kwargs) are just a straight pass-through to the handler.) However, in certain cases it is very useful to have access to the origin function return value. A common case is where the origin function creates a new object. The framework handles this internally by introspecting the handler function, and looking for **kwargs.

This is best illustrated with an example:

def origin_func(arg1: int, arg2: int) -> int:
    return arg1 + arg2

def handle_func1(arg1, arg2):
    # this func will not receive the return_value, as
    # no kwargs are specified

def handle_func1(arg1, arg2, **kwargs):
    # this func will receive the return_value via **kwargs
    assert "return_value" in kwargs

def handle_func1(arg1, arg2, return_value=None):
    # this func will receive the return_value

def handle_func1(arg1, arg2, return_value):
    # this func will receive the return_value, as it is a named arg,
    # and there is no *args variable

def handle_func1(*args, return_value):
    # this func will *NOT* receive the return_value

Internally, the app maintains a registry of side-effects functions bound to origin functions using the text labels. The docstrings for all the bound functions can be grouped using these labels, and then be printed out using the management command display_side_effects:

$ ./ display_side_effects

This command prints out the first line from the docstrings of all functions
registered using the @is_side_effect decorator, grouped by label.


    - Update CRM system.
    - Notify account managers.


    - Send confirmation email to user.
    - Notify customer service.

If you have a lot of side-effects wired up, you can filter the list by the label:

$ ./ display_side_effects --label update_profile

    - Update CRM system.
    - Notify account managers.

Or by a partial match on the event label:

$ ./ display_side_effects --label-contains profile

    - Update CRM system.
    - Notify account managers.

If you want to enforce docstrings on side-effect functions, then you can use the –check-docstrings option, which will exit with a non-zero exit code if any docstrings are missing. This can be used as part of a CI process, failing any build that does not have all its functions documented. (The exit code is the count of functions without docstrings).

$ ./ display_side_effects --check-docstrings

    *** DOCSTRING MISSING: update_crm ***
    - Notify account managers.

ERROR: InvocationError for command '...' (exited with code 1)

Why not use signals?

The above solution probably looks extremely familiar - and it is very closely related to the built-in Django signals implementation. You could easily reproduce the output of this project using signals - this project is really just a formalisation of the way in which a signal-like pattern could be used to make your code clear and easy to document. The key differences are:

  1. Explicit statement that a function has side-effects

  2. A simpler binding mechanism (using text labels)

  3. (TODO) Async processing of receiver functions

It may well be that this project merges back in to the signals pattern in due course - at the moment we are still experimenting.


The project is available through PyPI as django-side-effects:

$ pip install django-side-effects

And the main package itself is just side_effects:

>>> from side_effects import decorators


The project has pretty good test coverage (>90%) and the tests themselves run through tox.

$ pip install tox
$ tox

If you want to run the tests manually, make sure you install the requirements, and Django.

$ pip install django==2.0  # your version goes here
$ tox

If you are hacking on the project, please keep coverage up.

NB If you implement side-effects in your project, you will most likely want to be able to turn off the side-effects when testing your own code (so that you are not actually sending emails, updating systems), but you also probably want to know that the side-effects events that you are expecting are fired.

The following code snippet shows how to use the disable_side_effects context manager, which returns a list of all the side-effects events that are fired. There is a matching function decorator, which will append the events list as an arg to the decorated function, in the same manner that unittest.mock.patch does.

from side_effects import decorators, registry

def foo():

def test_foo():

    # to disable side-effects temporarily, use decorator
    with registry.disable_side_effects() as events:
        assert events == ['do_foo']
        assert events == ['do_foo', 'do_foo']

# events list is added to the test function as an arg
def test_foo_without_side_effects(events: list[str]):
    assert events == ['do_foo']

In addition to these testing tools there is a universal ‘kill-switch’ which can be set using the env var SIDE_EFFECTS_TEST_MODE=True. This will completely disable all side-effects events. It is a useful tool when you are migrating a project over to the side_effects pattern - as it can highlight where existing tests are relying on side-effects from firing. Use with caution.


Standard GH rules apply: clone the repo to your own account, create a branch, make sure you update the tests, and submit a pull request.


We are using it at YunoJuno, but ‘caveat emptor’. It does what we need it to do right now, and we will extend it as we evolve. If you need or want additional features, get involved :-).

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

django_side_effects-2.1.tar.gz (17.0 kB view hashes)

Uploaded source

Built Distribution

django_side_effects-2.1-py3-none-any.whl (14.4 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page