3.1 Redis Thread Model

This is a historical learning note and may contain outdated or incomplete understanding.

1. Redis Reactor Model

It consists of four parts: multiple sockets, I/O multiplexing, the file-event dispatcher, and event handlers.
The file-event dispatcher’s consumption queue is single-threaded, which is why Redis is called a single-threaded model.
- The Redis server opens a server socket to listen for client connections and registers this socket with the I/O multiplexer.
- After a client connects, the I/O multiplexer detects the accept event and pushes it into the queue; the file-event dispatcher gives the event to the accept handler, which creates a new socket and registers it with the I/O multiplexer.
- When the client requests data, the I/O multiplexer detects the read event and pushes it into the queue; the file-event dispatcher gives the event to the read handler, which reads the data, parses it into a command, and pushes it into the queue; the file-event dispatcher gives the event to the command-request handler, which processes it and pushes the result into the queue; the file-event dispatcher gives the event to the command-reply handler, which writes it to the socket.
- To respond with client data, the I/O multiplexer detects the write event and writes back to the client.

Redis has a single-threaded architecture, and command execution is completed in this thread.

A single thread means no thread-context switching and no locking problems for concurrent reads and writes.

First, Redis is an in-memory database, so all operations are naturally fast because they occur in memory.
Second, although it is single-threaded, it uses an I/O multiplexing model, so even one thread can serve multiple clients.
A single thread also means no thread-context switching.

Multi-threaded I/O is only used to handle network-data reads/writes and protocol parsing; command execution remains single-threaded.