metastable failure

RU: метастабильный сбой

A self-sustaining system collapse where a transient trigger (e.g. a traffic spike or a cache flush) activates a positive-feedback loop — retries amplify load, which causes more timeouts, which cause more retries — that persists long after the original trigger is gone. Distinguishing from ordinary overload: the system degrades past a tipping point from which it cannot recover without external intervention such as load shedding or traffic throttling.