How Airbnb Runs Distributed Databases on Kubernetes at Scale

Distributed SQL databases are not magic boxes that will handle availability and reliability challenges without thoughtful engineering. Airbnb’s case study is a great example of such engineering: don’t let more than one node replacement run at a time (serialized replacement controlled by custom k8s controller logic), spread a database across data centers (AWS availability zones), and don’t store data on the same node as DB server (detachable AWS EBS volumes).