Redis Core Principles Explained with Four Diagrams

Redis is often praised for being fast because it runs on a single thread, but the real reasons lie in three key techniques: all data resides in memory, I/O multiplexing with epoll handles thousands of connections, and efficient data structures such as hash tables (O(1) lookups) and skip lists for ordered traversal. Official tests show single‑threaded QPS exceeding 100,000. Since Redis 6.0, multi‑threaded I/O speeds up network send/receive, while command execution remains single‑threaded, a nuance many overlook.

The Append‑Only File (AOF) persistence works opposite to MySQL’s write‑ahead log: Redis first executes the command in memory, then appends it to the log. This design records only successfully executed commands, avoids syntax‑check overhead, and does not block the current write, but it can lose the last command if the server crashes and incurs a brief pause when the main thread flushes to disk.

RDB snapshot persistence creates a point‑in‑time binary dump of the in‑memory dataset. To avoid blocking writes during snapshotting, Redis forks a child process; the child writes the snapshot while the parent continues handling requests, leveraging the operating system’s copy‑on‑write (COW) mechanism. The save command runs in the main thread and blocks, whereas bgsave is the recommended non‑blocking approach.

High availability in Redis has evolved through three tiers. Tier 1 is master‑slave replication, providing read/write separation and data backup but requiring manual failover. Tier 2 adds Sentinel nodes that monitor masters, automatically elect a new master on failure, and perform unattended failover. Tier 3 introduces Redis Cluster, a decentralized architecture that hashes keys into 16,384 slots distributed across multiple instances, enabling horizontal scaling and full read/write performance, and is the dominant production solution today.

Summarizing the knowledge framework: thread model (pre‑4.0 single thread, post‑4.0 async deletion, post‑6.0 multi‑threaded I/O), persistence options (AOF, RDB with COW, hybrid), and high‑availability progression (replication → Sentinel → Cluster). Understanding these three dimensions gives both breadth and depth to Redis architecture.