Presentation: Netflix Play API - An Evolutionary Architecture

Suudhan: I work on the Playback API team whose identity is to deliver Playback Functionality 24/7. I, along with a team of stunning colleagues, own and operate the critical Play API service which orchestrates playback functions like deciding the best playback experience, authorize every playback and collect playback data for business intelligence. For the past 2 years, i lead the initiative to re-architect this service to significantly improve our scalability, availability and developer velocity.

Suudhan: Our API service have gone through 3 architectures. With Netflix scale, we often hit limitations of our architecture every 3 or so years. In this third iteration, we have architected a solution to optimize for evolvability. We fully expect things to change in 3-5 years time and we want an architecture in which each aspect of the architecture can be replaced with minimal overhead. My motivation for the talk is to share our learnings with this architecture and have an exchange of ideas with the qCon audience. I am particularly interested in hearing opinions about aspects which we might have overlooked or how some elements of our architecture can be applied to different domains.

Suudhan: Target audience is a senior software engineer who has had some experience with owning and operating services. The talk will appeal to persons who are thinking about re-architecting any part of their services; or starting afresh on a new architecture.

Suudhan: These are some of the problems i am thinking about:

1) How to convert real-time data services to materialized views to significantly improve our throughput and reduce point of failures in our critical systems.

2) How to isolate request-response style APIs from fire and forget ones to optimize for different attributes (like availability for the former and consistency for the latter)

3) When we see an availability drop, how to shed load by smartly degrading our user experience till the issue is resolved