GitHub's git repository infrastructure has grown very organically over the years, from a single testing server to today's sharded set of dozens of fileservers. As we outgrew the original system, it became very painful to develop new git features. We decided to start from scratch and design a system that fit the new way we store and manage git repositories. How do you replace the whole backend of GitHub, without making it impossible to work on, or impacting the lives of our millions of users?

In this talk I'll give a tour of where we came from, where we're heading, and how we're getting there.