Migrating thredUP Infrastructure to Kubernetes with HAProxy

ThredUP is the largest online consignment store for women’s and children’s clothes. ThredUP was launched in 2009 as a monolithic application running on Amazon Web Services. Since then it has evolved into a fully containerized distributed system powered by Kubernetes, with HAProxy playing a key role in routing nearly all of ThredUP’s traffic. Our migration resulted in faster releases with cycle times under 20 minutes and 60% lower hardware costs. We now support twice as many services, have unified infrastructure automation and have decreased new service rollout times from weeks to hours or minutes. This talk will describe the path we took to migrate to Kubernetes and how we used weighted and cookie-based load balancing to achieve our goals. We will talk about the issues we encountered and how we managed them.