Is Redis Single‑Threaded? Uncover Its Core Architecture and Multi‑Threading Features

Redis is often asked whether it is single‑threaded or multi‑threaded, single‑process or not; this article provides a comprehensive explanation.

Redis uses a dedicated thread to handle the majority of tasks, including data read/write, but recent versions also include multi‑threaded capabilities for certain operations.

Single‑threaded core

Redis’s single‑threaded data processing is efficient because it leverages I/O multiplexing, allowing one thread to monitor multiple I/O channels such as network connections and file descriptors, thus handling many client requests concurrently.

I/O multiplexing mechanisms

Redis relies on operating‑system system calls like select, poll, and epoll to listen to multiple client sockets.

select – the earliest mechanism, monitors a fixed set of descriptors but degrades with many descriptors.

poll – similar to select but uses a list, supporting more descriptors.

epoll – Linux’s event‑driven mechanism, avoids traversing all descriptors on each call, offering superior performance for high‑concurrency scenarios; Redis prefers epoll on Linux.

Because most Redis operations are in‑memory, combining them with I/O multiplexing minimizes I/O wait and yields high‑performance data handling.

Redis does employ multiple threads in special cases, such as persistence (RDB, AOF) where a child process handles disk writes, and since Redis 6.0 it adds optional multi‑threaded networking for read/write of network requests, while data processing remains single‑threaded.

Illustration of the architecture:

Persistence workflow example: