Presentation: Building Resilience in Production Migrations

I lead Billing Infrastructure Engineering at Netflix. We build the infrastructure that helps Netflix collect charges from its members. Part of that is to determine who should be charged and how much through our systems. We also hold all the gift codes and balances and track them. We also support major customer workflows. Our services participate in gift redemptions, Method of payment updates, cancellation requests etc.  We also do financial reporting.

To get a bit of why resilience is important specifically for us. We believe that for systems like us, we do our best when we are not visible, when nobody notices. It's a topic very close to my heart, how to build resilient systems, how to not go down. Our infrastructure  shouldn't be the reason people cannot watch what they wanted to. The motivation of the talk is primarily to share some of the things that we have seen over the years. Since I joined Netflix we've done major overhaul and nobody has noticed that infrastructure has been shifting. These kind of  projects have a thought process which starts even before we start the project, how to manage the resiliency, not going down. We have some interesting stuff to share that other people can benefit from as well.

There are two aspects when rewriting a system. One is not taking a downtime for the deployment of your code. And second is not taking a downtime to migrate the state from one store to other store. The talk is about how do you do these steps without being noticed. And we will share the stories of how we migrated at scale, the things we did, what helped us. How did we build confidence into our migration?,. How should you think about it when you want to not really impact customers, the different aspects of that.

My goal is for people to learn some of the tools and technologies that we have used, if they can find something similar in their world, and how possibly they can break down these bigger tasks into smaller buckets and be able to focus and solve it. Because every system that's running today around the world will need a rewrite at some point. I want people to walk away with confidence from our experience to solve a very common but complex problem in software. I'm hoping they'll learn some tips and tricks.