Design and Architecture of Flash Sale (秒杀) Systems

The article discusses the design of flash‑sale (秒杀) scenarios, where millions of users compete for a limited number of items in a very short time window, requiring strict data consistency and high performance.

Key problems identified include synchronizing client countdown timers, preventing bot or automated purchases, and ensuring backend servers can sustain the massive traffic.

Two main solution approaches are proposed. The first leverages CDN edge nodes to offload static resources, distribute lightweight services that collect online user counts, and assign request probabilities based on weighted user numbers, thereby filtering most traffic before reaching the data center.

The second approach combines CDN distribution, authentication to filter bots, load balancing (LVS + Keepalived) to spread requests across an Nginx cluster, followed by rate limiting, service degradation, and caching layers with pre‑warming and scheduled tasks to maintain database‑cache consistency and handle order processing.

The conclusion emphasizes that using edge‑node solutions and layered request filtering can effectively support flash‑sale traffic, while for broader high‑concurrency events like Double‑11, a more comprehensive high‑availability architecture with extensive performance testing and horizontal scaling is required.