I am a systems engineer at Allegro.pl, where I work on everything related to http, content serving, systems orchestration. I am a big fan of Linux. In my free time, I like to play with IoT, amateur radio and wireless networks. Outside technology, I enjoy travelling by bike with my friends.
With over 12 million of active users and up to 20Gbps traffic to serve, ensuring the stability and speed of the Allegro platform requires a reliable caching layer. Having the right tool do the right job is crucial. For that reason, we use both Varnish and nginx in our reverse caching proxy clusters depending on what we want to achieve. In our presentation, we will talk about how we route traffic and how we perform operations on HTTP headers and why we sometimes use SQL in VCL. We will talk about some of the tools we develop internally and about our open source Varnish cluster management tool called VaaS. We will also spend some time describing how we monitor our caching infrastructure. On the nginx side of things, we will demonstrate how things like trim, page cache, sendfile affect caching and why they matter. We will tell you how we reduced our image caching farm by half to serve twice as much traffic by choosing the right solution. We will also demonstrate how to feed empty cache quickly without impacting backend performance.