Redis clustering

A single Redis node fits in one machine's RAM. When you outgrow that, you shard with Redis Cluster: the key space is split into 16,384 hash slots, each node owns a subset of slots, and a key's slot is determined by CRC16(key) mod 16384. Clients connect to any node, which either serves the request or redirects (MOVED) to the node owning the key's slot. Two gotchas: multi-key operations across slots fail — you can't atomically operate on two keys unless they hash to the same slot, which is why "hash tags" exist ({user:123}.profile and {user:123}.cart hash together). Replication is per-shard — each master has replicas, automatic failover promotes a replica if a master dies, but writes during the election are lost. If you don't actually need the data volume, a single primary + replicas (Redis Sentinel for HA) is simpler than a cluster and solves most real problems. Cluster is for when memory genuinely exceeds one box.