Supporting code and slides for a talk originally given at DjangoCon-US, September 2010 (in Portland, Oregon, USA).

A tour through a few common "multiple database" usage patterns and how they can be implemented and utilized with Django. We'll talk about the strengths and weaknesses of each pattern and why you might not need any of them.

There are a few good reasons a system might want to interact regularly with multiple databases. “Because it’s what cool people do” is not one of those reasons. Most multi-database usages often fall into one of three main classes:

Separation by function.

All data of one type in one database, all data of another type in some other database.

Data replication (separation by access)

Some pieces of data are synchronized to multiple machines. Writes might go into one or more masters and reads normally come from the slaves.

Separation by sharding.

Data of a particular type (e.g. user records) is split across multiple databases, each database holding a shard of the whole data.

Obviously, combinations of these classes are possible, such as replicated sharded data in a huge site. There are tricks and traps to the way a developer talks to each of these sorts of setups. I'll spend a few minutes showing credible examples of the usage of each as well as when you might be over-engineering by going that way. All three access patterns are possible in Django 1.2, with varying degrees of ease of use and I'll show the type of code required in each case.

Each of the directories below this one has sample code for one of the above situations. Each case is a self-contained Django project. Follow the individual README.rst instructions to create and populate the databases in each case. By default, I am using the SQLite, but there is nothing database-specific about the code, so feel free to change the ENGINE setting for one or more databases and see what happens (there's no requirement in Django that each database uses the same backend).

If you want to browse / test the code in order of increasing complexity, I would recommend the following order:

• functional_split
• access_split
• sharding

Good Luck!

Malcolm Tredinnick (Sydney, Australia)