Table of Contents
There are three system management methods that greatly depend on the tools that you are using for managing these systems
Setting up and updating a system - be it a workstation or a server -
using manual installation, such as setting a new Mac, or by automated
methods such as cloud-init for CoreOS, constitutes a divergent system.
It generally implies bad management of a system, as it diverge from the
baseline overtime and is hard to replicate in case of hardware or
software failure.
Convergence is the process of mutating a divergent live system back to the baseline by comparing the state of the system to what the state of the system should look like and performing automated actions to get the system back to the baseline.
Puppet, Chef, Ansible and other similar tools follow the convergence method of maintaining systems. These tools are also considered declarative as they describe exactly how the system should look like.
Congruence is the practice of maintaining live hosts in complete compliance with a declarative baseline of the system. In other words, a congruent system is a system that you cannot alter without altering the representation in order to make the changes that you are attempting.
Working with a declarative baseline for an infrastructure (whatever the indented use case of the machine or machines in your network may be) allows the system to be replicated in a matter of minutes with no manual configuration or missed steps. It also allows you to know, at any point, what's the state of your entire infrastructure by simply reading the declarative baseline of it.
There's an excellent paper by Steve Traugott and Lance Brown published in 2002 that goes in details on all three methods.
My hopes with this project is to eliminate the constant need for manual intervention in order to fix one of my broken machine, for whatever the reason they broke. Even in the case of hardware failure, all I have to do is to replace the hardware knowing full well that my entire system can get back online by simply applying the baseline to a fresh machine and restore the user data only.