unexpected/non-linear bottlenecks, intermittent failures, cascading failures There is an architectural invariant which avoids these, which is that you don't bu

unexpected/non-linear bottlenecks, intermittent failures, cascading failures There is an architectural invariant which avoids these, which is that you don't build systems which can't tolerate the sudden loss of the machine they are running on. Actually that is not the invariant you were looking for. I was pointing specifically at issues that have little to do with machine failure but are frequent root-causes for system failure.