[ants_everywhere]

> it's not really possible to reduce the entire model down to "current state, desired state, next action."

This is basically how control theory works in general though. You have a state, a goal, and a perturbation toward the goal. I think this is the right level of abstraction if you want a powerful and flexible tool.

> it's not really possible to know how the various components will interact....Introducing a VDOM / reducer paradigm (like react) would go a long way towards solving these problems.

I think the problem here is that the physical characteristics and layouts of the machines makes such a huge difference that it would be prohibitively costly to virtualize or simulate this in a meaningful way. So instead, people use subsets of the physical structure to verify that configuration states work. You do this by having `dev`, `staging`, `prod` environments, using colored deployments, canary analysis, partial rollouts etc.

[lucianbr]

> I think this is the right level of abstraction if you want a powerful and flexible tool.

This says nothing about ease of use. And for software development, ease of use matters. Otherwise we would all use assembler, or at most C++. They're very powerful and flexible.

If anything, too much power and flexibility is a problem.