Article / 14th Sep 2008

Denormalisation Follies

One of the topics that popped up repeatedly at DjangoCon last weekend was how bad purely normalised table structures are, and how denormalisation is good for many things, including making your database cry less.

To that end, during the gaps in PyConUK this weekend, I decided to see how easy it would be to write a new Django field that will automatically denormalise a field in a related table across to another model.

Let's take an example; imagine we have this kind of model:

class Picture(models.model):
    user = models.ForeignKey(User)
    user_username = DenormField(User, "username")
    image = models.ImageField(upload_to="foo")
    title = models.CharField(max_length=100)

The idea is that the user's username (which, let's face it, you're going to need almost every time you need the photo) is stored in the same table, saving you from doing that join to get it from users each time. While this may not seem like much, it can be useful, and besides, it was fun to write.

My DenormField simply listens to a trio of the relevant signals, and then updates this value whenever it detects that the related User object has been updated, so it only costs on write (which, let's face it, is much better than on every read). You can do this kind of copying in normal Django app code as well, but I think this solution is much cleaner. I would do, I guess, since I wrote it.

It's only really a little test exercise in writing Strange New Fields (and I wanted to see how much monkeypatching it would take - none, as it turns out, just some very nasty code tricks). You can get a copy of it if you feel like having a play (warning: only briefly tested; may attach captions to your cat).