Back in February of 2012, I wrote an article titled 'Starting a Django Project the Right Way', and later followed up with 'Starting a Django 1.4 Project the Right Way'. Both of these articles still draw a consistent audience and are referenced in numerous StackOverflow answers, corporate wikis, and tweets. With 1.5 and 1.6 already out, now seems like an appropriate time to update the article again.
The beginning of a project is a critical time, when choices are made that have long term consequences. There are a number of tutorials about how to get started with the Django framework, but few that discuss how to use Django in a professional way, using industry accepted best practices to make sure your project's development practices scale as your application grows. A small bit of planning goes a long way towards making your life (and the lives of any coworkers) easier in the future.
By the end of this post, you will have
- A fully functional Django 1.6 project
- All resources under source control (with git or Mercurial)
- Automated regression and unit testing (using the unittest library)
- An environment independent install of your project (using virtualenv)
- Automated deployment and testing (using Fabric)
- Automatic database migrations (using South)
- A development work flow that scales with your site.
None of these steps, except for perhaps the first, are covered in the official tutorial. They should be. If you're looking to start a new, production ready Django 1.6 project, look no further.
Preparing To Install
I'm assuming you have Python installed. If you don't head over to python.org and find the install instructions for your architecture/os. I'll be running on a 64-bit Arch server installation hosted by Linode, with whom I'm very happy.
So, what's the first step? Install Django, right? Not quite. One common problem with installing packages directly to your current site-packages area is that, if you have more than one project or use Python on your machine for things other than Django, you may run into dependency issues between your applications and the installed packages. For this reason, we'll be using virtualenv and the excellent extension virtualenvwrapper to manage our Django installation. This is common, recommended practice among Python and Django users alike.
If you're using pip to install packages (and I can't see why you wouldn't), you can get both virtualenv and virtualenvwrapper by simply installing the latter.
After it's installed, add the following lines to your shell's start-up file (.zshrc, .bashrc, .profile, etc).
1 2 3
Reload your start up file (e.g.
source .zshrc) and you're ready to go.
Creating a New Environment
Creating a virtual environment is simple. Just type
django_project is whatever name you give to your project.
You'll notice a few things happen right away:
- Your shell is prepended by
- distribute and pip were automatically installed
This is an extremely helpful part of virtualenvwrapper: it automatically
prepares your environment in a way that lets you start installing packages using
pip right away. The
django_project portion is a reminder that you're using a
virtualenv instead of your system's Python installation. To exit the virtual
environment, simply type
deactivate. When you want to resume work on your
project, it's as easy as
workon django_project. Note that unlike the vanilla
virtualenv tool, where you run these commands doesn't matter.
"Wait, 'Installing Django'? I already have Django installed!" Fantastic. You aren't going to use it. Instead, we'll use one managed by virtualenv that can't be messed up by other users (or yourself) working elsewhere on the machine. To install Django under virtualenv, just type:
This should give you the latest version of Django which will be installed in your virtualenv area. You can confirm this by doing:
Which should point to your
$HOME/.virtualenvs/ directory. If it doesn't,
make sure you see
django_project before your prompt. If you don't, activate
the virtualenv using
Setting Up The Project
Before we actually start the project, we need to have a little talk. I've consulted on a number of Django/Python projects and spoken to numerous developers in the last few years. Overwhelmingly, the ones having the most difficulty are those that do not use any form of version control. It may sound unbelievable (considering the popularity of GitHub), but developers have simply never been exposed to version control. Others think that since "this is a small project," that it's not necessary. Wrong.
None of the tools listed here will pay greater dividends then the use of a version control system.
Previously, I only mentioned git as a (D)VCS. However, this project being in Python, Mercurial is a worthy Python based alternative. Both are popular enough that learning resources abound online. Make sure you have either git or Mercurial installed. Both are almost certainly available via your distro's packaging system.
If you plan on using git, GitHub is an obvious choice for keeping a remote repository. With Mercurial, Atlassian's Bitbucket is a fine choice (it supports git as well, so you could use it in either case).
(source) Controlling Your Environment
Even though we haven't actually done anything yet, we know we're going to
want everything under source control. We have two types of 'things' we're going
to be committing: our code itself (including templates, etc) and supporting
files like database fixtures, South migrations (more on that later), and a
requirements.txt file, which lists all of the packages your project depends on
and allows automated construction of environments (without your having to
install everything again).
Let's go ahead and create our project directory. Use the
command supplied by
django-admin.py to get it set up.
We'll see a single directory created:
django_project. Within the
django_project directory, we'll see another
containing the usual suspects:
wsgi.py. At the same
level as the second
django_project directory is
Intermezzo: Projects vs. Apps
You may be wondering why, back in Django 1.4, the
startproject command was added alongside the
startapp command. The answer lies in the difference between
Django "projects" and Django "apps". Briefly, a project is an entire web site or
application. An "app" is a small, (hopefully) self-contained Django application
that can be used in any Django project. If you're building a blogging application
called "Super Blogger", then "Super Blogger" is your Django project. If "Super Blogger" supports
reader polls, "polls" would be an Django app used by "Super Blogger". The idea is that
your polls app should reusable in any Django project requiring
polls, not just within "Super Blogger". A project is a collection of apps, along
with project specific logic. An app can be used in multiple projects.
While your natural inclination might be to include a lot of "Super Blogger" specific code and information within your "polls" app, avoiding this has a number of benefits. Based on the principle of loose coupling, writing your apps as standalone entities prevents design decisions and bugs in your project directly affecting your app. It also means that, if you wanted to, you could pass of the development of any of your apps to another developer without them needing to access or make changes to your main project.
Like many things in software development, it takes a bit of effort up-front but pays huge dividends later.
Setting Up Our Repos
Since we have some "code" in our project now (really just some stock scripts and empty config files, but bear with me), now is as good a time as any to initialize our repositories in source control. Here's how to do that in git and Mercurial.
This creates a git repository in the current directory. Lets stage all of our files to git to be committed.
Now we actually commit them to our new repo:
This creates a Mercurial repository in the current directory. Lets stage all of our files to git to be committed.
Now we actually commit them to our new repo:
If you plan on using a service like GitHub or Bitbucket, now would be a good time to push to them.
Using South for Database Migrations
One of the most frustrating aspects of Django is managing changes to models and the associated changes to the database. With the help of South, you can realistically create an entire application without ever writing database specific code. Changes to your models are detected and automatically made in the database through a migration file that South creates. This lets you both migrate the database forward for your new change and backwards to undo a change or series of changes. It makes your life so much easier, it's a wonder it's not included in the Django distribution.
When to begin using South
In previous articles, I recommended using South from the very beginning of your
project. For relatively simple projects, this is fine. If, however, you have a
ton of models that are changing rapidly as you prototype, now is not the time to
use South. Rather, just blow away and re-create the database whenever you need
to. You can write scripts to populate the database with some test data and edit
them as needed. Once your models stop changing, however, make the move to South
ASAP. It's as easy as
./manage.py convert_to_south <app_name>.
Installation and Setup
Still in our virtualenv, install South like so:
We setup South by adding it to our
file for the project. Add that now, as well as your database settings
for the project, then run
python manage.py syncdb.
You'll be prompted for a superuser name and password (which you can go
ahead and enter). More importantly, South has setup the database with
the tables it needs.
You may have noticed that we just ran
syncdb without having adding an app to the project. We do this first so that South is installed from the beginning. All migrations to our own apps will be done using South, including the "initial" migration.
Since we've just made some pretty big changes, now would be a good time to commit. You should get used to committing frequently, as the more granular the commit, the more freedom you have in choosing something to revert to if things go wrong.
To commit, lets see what has changed.
1 2 3 4 5 6 7 8 9 10 11 12 13 14
1 2 3 4 5
With both git and Mercurial, you may notice files you don't ever want committed,
like the compiled Python .pyc files and vim swap .swp files above. To ignore
these files, create a
.hgignore file in your root
project directory and add a shell glob pattern to match files you don't want
to be tracked. For example, the contents of my file might be
Before we commit, we have one more piece of information to track: our installed Python packages. We want to track the name and version of the Python packages we're using so that we can seamlessly recreate our environment in our production area. Helpfully, pip has a command that does exactly what we need.
I piped the output to a file called
requirements.txt, which we'll add to
source control so we always have an updated list of what packages are being used.
Let's stage and commit our settings.py and requirements.txt files to be committed by running
As developers become more comfortable with Django and Python, they realize that
settings.py file is simply a Python script, and can thus be "programmed".
One common pattern is for the
settings.py file to be moved from the rather
curious project directory to a new directory called
config. Just be
aware you'll need to make a slight change to
manage.py to accommodate the move.
INSTALLED_APPScan quickly grow into a morass of
third-party packages, in house django apps, and project specific apps. I like to
INSTALLED_APPSinto three categories:
- DEFAULT_APPS: Django framework apps installed as part of the default Django install (like the admin)
- THIRD_PARTY_APPS: Like South
- LOCAL_APPS: The applications you create
This makes it much easier to see what third-party applications you're using and what is home-grown. Just remember to eventually have a line similar to the following:
Otherwise, Django will complain about not having
Creating Our App
manage.py to create an app in the normal way (
startapp myapp) and add it under
INSTALLED_APPS. Also, take the time to make
chmod +x manage.py) so you can just type
rather than needing to type
python manage.py <command> all the time. Honestly,
so few developers do this. I can't for the life of me figure out why.
The first thing we'll do, before adding models, is tell South we want South to manage changes to our models in the form of migrations:
This creates a migration file that can be used to apply our model changes (if we had any) to the database without needing to completely destroy and rebuild it. It also also allows us to revert changes if things go sideways on us. We use the migration file to migrate the database changes (even though there are none) using :
South is smart enough to know where to look for migration files, as well as remember the last migration we did. You can specify individual migration files, but it's usually not necessary.
When we eventually make changes to our model, we ask South to create a migration using:
This will inspect the models in
myapp and automatically add, delete,
or modify the database tables accordingly. Changes can then be applied to the
database using the migrate command as above.
Our Development Area
A good habit to get into is to write and test your code separately from where
you're serving your files from, so that you don't accidentally bring down your
site via a coding error when you're adding new functionality, for example.
git and Mercurial make this simple. Just create a directory somewhere other than
django_project is installed for your development area (I just call it
In your development (
dev) directory, clone the current project using git or Mercurial:
Both tools will create an exact copy of the entire repository. All changes, branches, and history will be available here. From here on out, you should be working from your development directory.
Since branching with both git and Mercurial is so easy and cheap, create branches as you work on new, orthogonal changes to your site. Here's how to do it each tool:
Which will both create a new branch named
master mimics the "production" (or "version live on your site")
master and can be used for recovery at
Note that branching is kind of a contentious topic within the Mercurial
community, as there are a number of options available but no "obviously correct"
choice. Here, I use a named branch, which is probably the safest and most
informative style of branching. Any commits after the branch command are done on
Using Fabric for Deployment
So we have the makings of a Django application. How do we deploy it? Fabric. For a reasonable sized project, discussing anything else is a waste of time. Fabric can be used for a number of purposes, but it really shines in deployments.
Fabric expects a fabfile named
fabfile.py which defines all of the actions we
can take. Let's create that now. Put the following in
fabfile.py in your project's root directory.
1 2 3 4 5
This will run the tests and commit your changes, but only if your tests pass. At this point, a simple "pull" in your production area becomes your deployment. Lets add a bit more to actually deploy. Add this to your fabfile.py:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
This will pull your changes from the development master branch, run any migrations you've made, run your tests, and restart your web server. All in one simple command from the command line. If one of those steps fails, the script stops and reports what happened. Once you fix the issue, there is no need to run the steps manually. Since they're idempotent, you can simply rerun the deploy command and all will be well.
Note that the code above assumes you're developing on the same machine you
deploy on. If that's not the case, the file would be mostly the same but would
run function instead of
local. See the Fabric documentation for details.
So now that we have our
fabfile.py created, how do we actually deploy?
Simple. Just run:
Technically, these could be combined into a single command, but I find it's better to explicitly prepare your deployment and then deploy as it makes you focus a bit more on what you're doing.
Setting Up Unit Tests
If you know anything about me, you probably know I'm crazy about automated
tests. Too many Django projects are written without any tests whatsoever. This
is another one of those things that costs a bit of time up-front but pays
enormous dividends down the road. If you've ever found yourself debugging
your app using
For Django, the Python
unittest module is perfectly sufficient. The following
is a minimal example of tests for a single app:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
You would put this code in a file called
test_<appname>.py and place it in the
same directory as the app it is testing. To run the tests for an app, simply run
./manage.py test <appname>. The fabfile we created already knows to run the
tests before deployment, so no need to make any other changes.
Enjoy Your New Django Application
That's it! You're ready to start your actual development. Now is when the real fun begins. Just remember: commit often, test everything, and don't write code where you serve it from. Regardless of what happens from here on out, you've definitely started a Django 1.6 project the right way!Posted on by Jeff Knupp