Running Django with Gunicorn - Best Practice

Question

There are 3 ways to run a django application with gunicorn:

1. Standard gunicorn + wsgi (ref django doc)

   gunicorn project.wsgi:application

2. Using gunicorn django integration (ref gunicorn doc and django doc):

   python manage.py run_gunicorn

3. Using gunicorn_django command (ref gunicorn doc)

   gunicorn_django [OPTIONS] [SETTINGS_PATH]

Django's documentation suggests using 1., which is not even listed as an option on Gunicorn documentation.

Is there any best practice on the best way to run a django app with gunicorn, and what are the foreseable advantages/disadvantages of these different solutions?

Taking a glimpse at gunicorn's code it looks like they pretty much all do the same: 2. seems to be creating a wsgi app using django's internals, and 3. uses 2.

If that's the case, I wouldn't even understand what's the reason for not simply using "1." all the time, especially since a wsgi.py file is autocreated for you since django 1.4; if that's true maybe simply a documentation improvement should be suggested...

Also, best practice for gunicorn settings with django would be great. Using 1., does it make sense to set some defaults in the wsgi file and avoid additional settings?

References:

1. Should I use django-gunicorn integration or wsgi? only concerns choices 1. and 3., there's no hint for the settings and the answer gives no rationale
2. Deploying Django with gunicorn and nginx give some broader information but is not strictly related nor answer this question
3. Django Gunicorn wsgi about version "4", which is launching gunicorn -c configfile and configfile will point to django_settings to django
4. Django WSGI and Gunicorn is just a bit confusing :) mixing up 1. and 3. Of course wsgi.py is used only with 1.

Answer 1

After checking out I'd say that the best way is using gunicorn + wsgi

$ gunicorn project.wsgi:application 

It's now both confirmed in gunicorn docs: if you run Django 1.4 or newer, it’s highly recommended to simply run your application with the WSGI interface using the gunicorn command and django as linked above.

It also avoids adding gunicorn as installed app, which means it's not a requirement to install gunicorn to test your app which might be useful from time to time.

About Settings

The Django settings file to be used can be passed through an ENV variable, or customized in the wsgi.py file. I sometimes create several wsgi.py files if I have multiple settings (eg. multiple websites) that have to run from the same project - See Django Doc for more info.

A one-liner solution that does not require any new file from Carl's comment:

DJANGO_SETTINGS_MODULE=project.settings.prod gunicorn project.wsgi:application

sounds like a nicer way (though I'll probably end up writing it in some shell commands to make it easy to "remember").

Gunicorn settings can be passed as -c settings_file, but I'm exploring other ways and will try to update this answer if I find any. Using environment variables seems a workaroud, but only for limited cases

In particular it would be nice to get/share some settings between django and gunicorn; gunicorn documentation says:

Currently, only Paster applications have access to framework specific settings. If you have ideas for providing settings to WSGI applications or pulling information from Django’s settings.py feel free to open an issue to let us know.

(Update: haven't found any smarter way, but after all env variables are enough for my most-common cases).

Answer 2

I guess using run_gunicorn is the way to go, it's also the simplest way to use it.

It's basically the same as usign gunicorn project.wsgi:application but needs gunicorn to be added to INSTALLED_APPS so that django recognizes the run_gunicorn command, therefore it's probably not the default way...

Using gunicorn_django is more or less deprecated, as the documentation also states here...