How Amazon RDS Replication Works and Why the FAA’s Database Problem...

Everyone claiming that "problem X could never occur on (our) system Y" should be approached with lots of scepticism

They either define "problem X" very narrowly (so that a slightly-different problem, X', could still happen and with essentially the same result as X), or they are overconfident in their system in general and just can't seem to imagine that things like (for example) "two datacenters burning down at the same time" could ever happen