Design of a High‑Concurrency Inventory Search System with Caching and Asynchronous Processing

Qunar's ticket inventory is a core business where each seat is managed as a stock item; the system must handle millions of daily search and booking requests across many channels.

The inventory search workflow includes supplier operations (stock entry, price adjustment, value‑added services) and user actions (price search, stock reservation, stock release). Core metrics are inventory turnover efficiency and accuracy.

Challenges arise from a large number of personalized channels, causing high request QPS (>5K) and the need for minute‑level price updates, leading to pressure on the inventory service.

Solution design focuses on three goals: reduce invalid requests, improve cache hit rate, and enhance cache update efficiency.

Reduce invalid requests : implement product cataloging to let channels filter requests locally and use channel cache notifications to keep channel caches fresh.

Improve cache hit rate : employ request replay to keep hot cache entries, and apply cache isolation to separate common attributes (cost price, available quantity) from channel‑specific data.

Enhance cache update efficiency : adopt cold‑hot segregation of replay queues, unified cache manager, and complete closed‑loop notifications for events such as stock addition, price change, and sale.

Additional techniques include a work‑shard exchange space for replay queues, full‑process asynchronous processing using Spring Web MVC async and MQ callbacks, and careful thread‑pool separation to achieve high throughput while maintaining real‑time response.

The design provides a reference for similar high‑concurrency inventory search systems.