How can a whole data center be described in a text file?

A whole data center can be described in a text file because the files tell machines exactly how to set themselves up. This is infrastructure as code: managing and provisioning data center resources through machine-readable definition files, instead of by hand. The same code can create one server or hundreds, the same way each time.

Martin Fowler remembers the old way as the Iron Age. To release a new server application you had to find physical hardware, configure it and deploy onto it, and getting the hardware usually took months. Now, in the Cloud Age, firing up a new server is a matter of seconds and needs only an internet connection and a credit card. Software commands create servers, set them up and tear them down.

Speed is not the main point, though. When people build servers by hand, different readings of imprecise instructions, let alone errors, produce snowflakes: servers with subtly different configurations that cause faults that are hard to debug. The rule of infrastructure as code is that nobody logs in to a server and makes on-the-fly adjustments. Every change goes into the code instead.

The code is then kept in source control, so every configuration and every change is recorded for audit, and it can be tested like any other program. Using configuration code makes changes safer, faults are found and fixed more quickly, and at worst changes can be reverted to the last working configuration. It's the same safety net programmers have for their own code, applied to the machines that run it.