Livelock

A livelock is a concurrency failure in which processes are not blocked — they continue to execute and communicate — but their execution produces no progress toward the system's goals. Unlike deadlock, where processes wait indefinitely for resources that will never be released, livelock is a dynamic trap: the processes are active, but their activity is structured so that each response triggers a counter-response that returns the system to its original state. The system is not frozen. It is running in place.

Livelock is common in protocols that use retry or back-off mechanisms without sufficient randomization. Two processes detect a collision, back off, retry simultaneously, collide again, and repeat forever. The protocol is correct in every local step but incorrect globally. This is the hallmark of livelock: local correctness without global progress. The consensus literature has produced many livelocked systems — protocols that preserve safety at the expense of liveness, responding to every failure with a recovery that generates a new failure.

The lesson of livelock is that activity is not progress. A system that is busy is not necessarily a system that is working. The appropriate response to livelock is not more concurrency but better coordination — or, paradoxically, less.