This talk will dive into how we rewrote one of our production services in Go, leveraging Golang’s natives proxy implementation and routines alongside its async capabilities for improved scale & throughput of web services, enabling exponentially improved performance.
AppsFlyer, a leading mobile attribution & marketing analytics platform, processes nearly 70+ billion HTTP requests a day (approximately 50 million requests a minute), and is built using a microservices architecture. The entry point to the system that wraps all of the frontend services is a mission-critical (non-micro) service called the API Gateway. This essentially serves as a single point for routing traffic from customers to our backend services, simplifying authentication and authorization exponentially for our clients, but with the tradeoff of also potentially being a single point of failure.
Originally, this service was written in Clojure. As traffic grew - it became apparent that the code for the API gateway was too complex, and needed constant refactoring to enable the throughput required. Once the service became too unstable, we realized the we needed to rewrite the project completely - either in Clojure (just better), or explore other language options as well. This project decided to forego cognitive biases - and explore new language to rewrite the service to. After benchmarking, Go was selected and then went through a rigorous design phase, then rewrite, migration of production services, and benchmarking for improved performance. This talk will walk you through how to qualify a new language to introduce for mission critical production services, best practices for rewriting and migrating production services.
Below is an outline of the talk structure:
* Brief intro to describe technology stack & scenario
* Previous architecture and need for rewrite
* Design, Implementation, Architecture