Redis Deep Dive: Core Technologies, Evolution, and Real-World Practices

Interview with Zhang Donghong, Redis China User Group Chair

The rapid growth of the Internet and Web has driven enterprises to handle ever‑larger data volumes and higher concurrency, prompting a shift from relational databases to NoSQL solutions such as Hadoop, Cassandra, MongoDB, and Redis.

Self‑Introduction and Team Focus

Zhang Donghong is a senior DBA at Sina Weibo R&D Center, leading a team responsible for database operations and service assurance across Weibo’s platform, mobile app, topics, red‑packet, open platform, private messages, and content‑control projects. The team manages MySQL, Redis, Memcached, MCQ, Kafka, Pika, PostgreSQL and other products.

Team Work Emphasis

Weibo’s massive user base (over 376 million monthly active users) generates unpredictable traffic spikes during events, requiring a robust, stable, and automated database operation system. The team builds an automated DBA platform, integrates ticket‑driven deployment, intelligent alarm classification, capacity planning, HA, elastic scaling, and real‑time monitoring.

Redis Version Evolution

Key improvements across major releases:

Redis 2.6 : millisecond key expiration, read‑only replicas, Lua scripting, removal of client‑connection hard limit, removal of virtual memory.

Redis 2.8 : incremental replication, clearer process name, pub/sub commands, Sentinel 2.0, configurable maxclients.

Redis 3.0 : Redis Cluster, optimized string encoding, LRU enhancements, configurable maxmemory units, CLIENT PAUSE.

Redis 3.2 : GEO support, quicklist, SDS optimizations, enhanced Lua, new RDB format, faster CLUSTER NODES.

Redis 4.0 : module system (RediSearch, ReJSON, Redis‑ML), psync2.0, LFU eviction, non‑blocking DEL / FLUSHALL / FLUSHDB, RDB‑AOF hybrid persistence, MEMORY command, NAT/Docker‑compatible Cluster, Jemalloc integration.

Data Types and Use Cases

Redis provides String, Hash, List, Set, Zset, Bitmap, HyperLogLog, GEO, etc. Typical scenarios include session cache, full‑page cache, SMS verification, rate limiting, message queues, pub/sub notifications, leaderboards, counters (likes, reads, posts), social graphs, push notifications, inventory control, real‑time financial metrics, UUID generation, and location‑based services.

Redis as a Message Queue

Two implementation methods:

List‑based queue – persistent, cluster‑aware, simple API. Pros: durability, clustering, ease of use. Cons: single consumer per message, manual HA handling, risk of memory overflow under fast producers.

Pub/Sub – multiple consumers, no persistence. Pros: many consumers, simple API, cluster support. Cons: fire‑and‑forget, no guarantee of delivery, no built‑in performance guarantees, possible memory pressure if consumers lag.

High‑Availability Solutions

Common HA patterns include:

Keepalive + Redis

Redis Sentinel

Twemproxy + Sentinel + Redis

Redis Cluster

Sentinel + Proxy + Zookeeper + Redis

Zookeeper + MySQL + Redis + DNS

These solutions address failover, load balancing, and cross‑data‑center disaster recovery.

Automation Challenges

The biggest hurdle is achieving “intelligent” automation (AIOps). Required capabilities include a self‑service DBA platform, one‑click deployment, intelligent alarm classification, auto‑scaling, real‑time monitoring dashboards, sub‑second HA switching, elastic cache scaling, and extensive scripting to simplify operations.

Future of Databases

Hardware advances (PCIe, 25 GbE) and cost reductions will drive more Redis‑based services, including Redis Enterprise, Cloud Redis, and open‑source NewSQL alternatives like Pika and TiDB. Modular extensions (RediSearch, Redis‑ML) will expand Redis into AI, IoT, blockchain, and big‑data ecosystems.

Advice for Beginners

Read the official documentation, follow industry blogs, study the few available books, practice hands‑on, explore source code, keep notes, and stay active in the community to learn from real‑world cases.

Final Thoughts

Keep up with Redis releases, explore the surrounding ecosystem (tools, middleware, alternatives), and consider broader database trends to stay ahead.