Firstly, there’s a little problem of terminology to take care of. What many people call a web application, Django calls a “project”. Instead, the Django team uses the term “application” to describe a web application component that can be deployed into one of many projects. To describe it in the wordpress paradigm, wordpress could be a Django project (if it weren’t written in php) but the blogging component, the tagging component and the themes manager might all be separate Django applications.
This distinction really pushes the concept of re-using components. For example, once some one were to write a tagging Django application, the same app could be deployed for photo tagging, blog tagging and other types of content management. These applications are supposed to be completely contained and include their seed data (fixtures), database models, and templates.
Packaging python modules is a relatively trivial task and there is a well defined approach for it: distutils. This creates a more or less standard installer that allows anyone to install your package with a single command. From there, it can be put into the python package index and easy_install can install your python package (and dependency packages) with a single command. This works great for python packages like Django and BeautifulSoup, but how would it work for a whole web application? It made me wonder how Ellington, the flagship Django product, does it.
When distributing a python module, it makes sense to support as many platforms and configurations as possible. However, when deploying a full web application into a production environment, it makes sense to restrict your platform to what is tried and true. Ellington’s website, for example states:
Ellington takes advantage of the most secure and flexible open-source technology available: Apache for web serving, Python for programming, PostgreSQL for data, all optimized to work together on a stable, high-performance Linux platform.
Ellington isn’t intended to run on a wide variety of platforms even though it probably could. It is meant to run a production grade newspaper and therefore they specify its exact dependencies — probably the specific versions of apache, postgres, python and even linux!
So what have I learned about deployment and what do I do with all the media? After browsing django-users and the blogosphere, sticking with distutils is a great idea. I think that packaging each separate Django application is good idea. Each package is completely self contained and includes its media, templates, and code. In addition, the project settings should be minimal and possibly contained in another easily deployed python module controlled by distutils. In terms of fitting the whole thing together and deploying end to end, this is where the native package manager comes into play. This is the best way to manage both python and external dependencies. Rpm, msi or deb installers could fetch all the appropriate python modules (your Django applications), install the right version of your database and web server, sync your database, create the symbolic links to your media and even fill out your basic settings. For larger installations that require the database to be split from the python code and from the static media, this process still makes sense with few changes.
One thought on “Deploying Django Powered Web Applications”
It would be unfair to talk about deployment without mentioning Fabric. Fabric combines the benefits of a native package manager, except that it is cross platform, with the ability to deploy on multiple hosts remotely. For those familiar with Rails, Fabric is like Capistrano.
Comments are closed.