Redis 클러스터 및 샤딩

A single Redis instance is limited by one server's memory and throughput. Redis Cluster spreads data across many nodes so the cache can grow far beyond a single machine.

Redis Cluster divides the keyspace into 16384 hash slots. Each node owns a range of slots. A key belongs to exactly one slot, determined by hashing the key.

The slot is computed as CRC16(key) mod 16384. The client (or cluster) uses this to know which node holds the key, so reads and writes go directly to the right place.

With three nodes, slots split roughly into thirds. Adding a node means moving some slots to it; removing one redistributes its slots. This is how the cluster rebalances.

If a client contacts the wrong node, the node replies with a MOVED redirect pointing to the correct node. During slot migration, a temporary ASK redirect is used. Smart clients cache the slot map.

Operations across multiple keys (like MGET or transactions) only work if all keys live in the same slot. Otherwise the cluster rejects the command, because the keys may be on different nodes.

Hash tags solve this. Only the part inside curly braces is hashed, so keys sharing a tag land in the same slot. user:{42}:profile and user:{42}:cart are co-located.

Each master node can have replicas. If a master fails, a replica is promoted automatically, so the slot range stays available. This combines sharding with high availability.

Before Redis Cluster, apps sharded manually by hashing keys to instances in client code. Cluster mode moves that logic into Redis itself, handling rebalancing and failover automatically.

Clustering adds power but also constraints: cross-slot operations are limited, multi-key Lua scripts need same-slot keys, and operations are more complex. Use it when a single node truly is not enough.

Reach for Redis Cluster when memory or throughput exceeds one node, when you need automatic failover at scale, or when a region demands many nodes. For smaller caches, a single primary with a replica is simpler.

You learned Redis Cluster and sharding:

• The keyspace splits into 16384 hash slots owned by nodes.
• CRC16 routing plus MOVED/ASK redirects locate keys.
• Multi-key ops need same-slot keys; hash tags co-locate them.
• Replicas add automatic failover.

Clustering scales a cache horizontally when one node is not enough.