It's a nice ideal to aim for. But I agree, it's not "must" and not "exactly". If I run into a "very-interesting-problem" which depends on synchronisation between multiple machines and includes some geographic dependencies and caching, I may do the mental calculation:

- reproducing the issue in a repeatable way would take a week or more

- I can make 2 educated-guess fixes a day which don't affect production negatively

I would be a fool to choose the first (or at least start with the first) in most companies. I mean, I want to display pictures on the internet, not send people into space.