Rollbacks and Immutable Infrastructure

When production breaks, rollback is only easy if the last good version still exists and you can deploy it quickly.

On servers that are changed in place, that is hard. People edit config files, upgrade packages, restart services, and apply small fixes over time. After a while, nobody is fully sure what changed. The "good version" is no longer a clean thing you can point to.

Immutable infrastructure avoids that problem. Each version is built once and stored as a deployable artifact, such as a container image or VM image. Servers are not patched in place. They are replaced with new ones.

Then rollback becomes much simpler: send traffic back to the previous artifact.

This chapter explains what immutable infrastructure means, how it makes rollback easier, when to roll back versus roll forward, and what can still go wrong.