The common and "traditional" way of doing distributed locking is with a coordination service like ZooKeeper. ZooKeeper style services have an advantage of no TT

The common and "traditional" way of doing distributed locking is with a coordination service like ZooKeeper. ZooKeeper style services have an advantage of no TTLs - the moment a process dies, the lock is released, and the next in line waiting on the lock is immediately notified. Redis/Memcache with a TTL serves this purpose for the most part, but if you require as close to a 100% guarantee that 1 and only 1 process holds the lock at any given time, these will eventually fail you. Think network partitions, tasks outlasting the TTL, replication lag/eventual consistency etc. ZooKeeper and similar use concensus protocols like ZAB, Paxos or Raft to provide guarantees even in the face of failure.