There are many considerations where the fault lines should be. Somethings absolutely need upstream support and cannot just be buffered… for example payment processing.

Shit happens and we DevOps people do everything we can to minimize it. That’s why your apps might go down for like 2hrs a year.