Rolling Deployments

A rolling deployment replaces the running version of an application incrementally. The system swaps out instances a batch at a time: new instances come up on the new version, old instances drain and shut down, and some part of the fleet keeps serving traffic at every moment.

Rolling is the default in Kubernetes, AWS ECS, Nomad, and almost every modern orchestrator. The simplicity hides some real depth, though: the application runs in a mixed-version state against the same database and load balancer, and most rolling-deployment failures come from how the system passes through that state.