Another comment on this thread pointed out they had a potential collision in their instance name generation which may have caused this. That would mean this was user error, not a reliability issue. AWS doesn’t require instance names to be unique.