I remember reading Facebooks caches had a dedicated standby set of “gutter” servers that would take over a failure quickly (otherwise inactive and unused) that

I remember reading Facebooks caches had a dedicated standby set of “gutter” servers that would take over a failure quickly (otherwise inactive and unused) that was an interesting mitigation for some failure scenarios.