Choosing and Implementing Distributed Cache Systems with Redis

Cache systems are essential for improving concurrency, throughput, and response speed, but single‑node solutions may become insufficient as data volume grows, necessitating distributed caching.

1. Cache System Selection

Cache types are broadly categorized into CDN cache, reverse‑proxy cache (e.g., Nginx), local cache (e.g., EhCache, Guava), and distributed cache.

CDN cache: edge nodes store data.

Reverse‑proxy cache: Nginx caching.

Local cache: EhCache, Guava.

Distributed cache: various systems.

The article focuses on five distributed caches, comparing Memcache, Tair, Redis, EvCache, and Aerospike. EvCache (Netflix) and Aerospike (SSD‑based NoSQL) have limited applicability.

Tair (Alibaba): cross‑data‑center, linear performance scaling, suitable for large data volumes. Engines: LDB (LevelDB‑based), MDB (memcache‑based), RDB (Redis‑based).

Memcache: limited synchronization and distributed support.

Redis: most active community and widely used.

2. Redis Cluster Cache Solutions

Three high‑availability approaches are discussed:

Master‑Slave Mechanism

Simple read/write separation with a master handling reads/writes and slaves providing backups; however, recovery is complex and scaling is limited.

Sentinel Mechanism

Redis Sentinel monitors masters and promotes slaves to master on failure, offering better automatic failover.

Distributed (Redis‑Cluster)

Redis‑Cluster provides a decentralized solution with 16384 hash slots, enabling linear scaling to thousands of nodes while sacrificing some consistency.

Key points include asynchronous replication, node‑to‑node ping for failure detection, and the need for keys involved in multi‑key operations to reside in the same slot (using hash tags).

3. Common Cache Issues

Cache penetration: handling requests for non‑existent data (empty values, Bloom filters).

Cache breakdown: preventing hot‑key expiration spikes (mutex locks, non‑expiring keys).

Cache avalanche: mitigating massive simultaneous expirations (randomized TTL, background refresh, rate limiting, dual‑cache).

Cache consistency: strategies such as Cache‑Aside, Write‑Back, Read‑Through/Write‑Through.

Hot data handling: key splitting, slot migration, multiple replicas.

Cache warm‑up and degradation techniques.

4. Redis Cluster Usage

To create a Redis cluster, run the following command on any node (replace IP and port placeholders with actual values):

redis-cli --cluster create IP1:port1 IP2:port2 IP3:port3 IP4:port4 IP5:port5 IP6:port6 ... --cluster-replicas 1

After the cluster is set up, use cluster node and cluster info to inspect status. Java developers can enable Redis Cluster in Spring Data Redis by configuring master nodes:

spring.redis.cluster.nodes=ip1:port1,ip2:port2,ip3:port3

Add the dependency:

compile("org.springframework.boot:spring-boot-starter-data-redis")

Then use RedisTemplate for operations.

5. Summary

The article starts from cache system selection, introduces several Redis‑based cluster solutions, highlights Redis Cluster as the optimal choice for most scenarios, enumerates common cache problems and remedies, and provides practical setup instructions for Java backend developers.