A single failed server is easy to deal with. Pull it, replace it, toss it in a bin of spares. A whole rack, a data hall,...
