MySQL was initially chosen as their database of choice because it is
free, simple, fast and reliable. All told, they have about 6000 server years of runtime experience without data loss or corruption due to software. However like most megascale services they often end up using MySQL as a glorified key<->value store as opposed to taking advantage of the relational features of the database. When asked why bother using a relational database, the response is that the database management features such as data replication and administration are still valuable even if you aren’t using foreign keys and other relational features. One interesting challenge is that they often use “data driven schemas” to enable programmers to add new types to the database without requiring schema changes. This approach sounds similar to what Bret Taylor described in his post
How FriendFeed uses MySQL to store schema-less data. The challenge with this approach is that you end up with key<->value pairs stuffed in some column in the database where you can’t even efficiently query because you can’t index the fields you care about.
