Didn’t AWS have a cascading failure last year because of a missing interlock on a script to take part of a data center offline? They discovered that it takes a lot longer than they remembered to reintroduce that many machines to the cluster. They came away from that incident with two or three action items.