Pinterest is a cloud first visual discovery engine that serves over 320MM users. Pinterest built its continuous delivery platform on Spinnaker. And today it’s used to deploy our services to VM hosts, K8S pods and real time data processing framework. In July 2020, Spinnaker averaged 800 deployments daily. In this talk, I will share how we designed our CD system to deploy itself. And the topics include: 1. Impact on Spinnaker’s availability & reliability Spinnaker is the foundation of our CD system, and its SLA has critical production impact. In the past, we deployed Spinnaker microservice pods via a command line tool and then manually tested the new deployment,which was prone to human error. I will present our new approach that automates the deployment and rollback of prod Spinnaker with a staging Spinnaker pipeline, which decreased deployment time, increased deployment frequency, and significantly reduced the oncall person’s time spent on deployment, as well as production incidents. 2. Challenge to deploy microservices with complex k8s configmap I will share how we built a Spinnaker custom stage to work with other components of Pinterest’s CD system (Hermez) to deploy to K8S.
