What does "immutability" mean in the context of infrastructure as code?

Systems are maintained entirely as code and never changed by hand. Any change is made in the code and the system is rebuilt or redeployed from it.

* A hand-patched server drifts; an immutable one is rebuilt from code, so rollback means deploying the previous version. *

On a mutable server, an admin logs in and edits a config file or installs a package. After a year of such fixes, nobody knows exactly what is on the machine, and no two "identical" servers really are identical. This is called configuration drift.

Immutable infrastructure forbids that. If something has to change, you change the code (a template, a container image, a playbook) and replace the running system with a freshly built one. The benefits:

• the running system always matches what is in the repository,
• any version can be rebuilt exactly, so rollback is just deploying the previous version,
• every change is visible in the version history.

Tip: Treat servers like cattle, not pets. A sick one is replaced, not nursed.

Go deeper:

• Martin Fowler — Immutable Server — the short essay that named the pattern.