Redis Sentinel
Redis HA solution without sharding. Sentinel processes (3+) monitor master + replicas, vote quorum on failover. Client asks Sentinel for current master address.
Simpler than Cluster when data fits on one master. Distinct from Cluster: Sentinel = HA only, Cluster = HA + sharding. Client libraries (Jedis, Lettuce, ioredis) have Sentinel-aware mode — abstracts discovery. Failover time 5-30 sec typically (configurable down-after-milliseconds + failover-timeout). Split-brain risk on network partition + quorum failure. Modern alternative: managed services (ElastiCache, MemoryDB) make Sentinel mechanics transparent.