How Alibaba Handles Million‑Ticket Flash Sales with Scalable Backend Architecture

Background

Ticket sales for movie festivals are extreme flash‑sale scenarios; for example, the Shanghai International Film Festival sold 50,000 tickets in one minute. Alibaba senior developer Nianxian describes how the entertainment cloud‑intelligence system supports such high‑traffic concurrency, guarantees stability, and avoids duplicate sales.

Architecture Design

The system is divided into three layers: channel access, business, and service. The business layer separates external services from admin functions, while the service layer provides stable foundational services.

Implementation and Technical Analysis

1. How to Resist High Concurrency

The solution consists of three parts: hotspot data isolation, flow‑shaping funnel, and multi‑level cache.

Hotspot Data Isolation separates inventory‑related hotspot data into a dedicated database with sharding, reducing pressure on the main system, and isolates business data by pre‑warming caches.

Flow‑Shaping Funnel applies layered request filtering (account security, purchase eligibility, venue status, product status, flash‑sale end) to drop invalid requests early, dramatically reducing write traffic to the database.

Multi‑Level Cache combines distributed cache (e.g., Tair) with a local cache (LocalCache) using three expiration strategies (2 s, 60 s, 122 s). Cache pre‑heating, expiration‑time control, and tiered reads (local first, then remote) protect the system from traffic spikes.

Local cache stores static data such as user, permission, and basic info.

Distributed cache stores full entity information (DO and VO) using Tair.

2. System Stability and High Availability

Stability is ensured through full‑link pressure testing, rate limiting, degradation, dynamic scaling, traffic scheduling, reducing single points of failure, and simplifying dependencies.

The preparation process includes:

Identifying weak points in architecture and core flows.

Conducting full‑link pressure tests to determine maximum QPS.

Configuring appropriate rate‑limit thresholds.

Creating emergency response plans.

Implementing strict account‑permission controls.

Running pre‑battle rehearsals.

Documenting a battle handbook.

During the battle, teams focus on monitoring, rapid response, and detailed recording of peak metrics. Post‑battle activities involve comprehensive review, issue analysis, and knowledge consolidation.

Best Practices include precise monitoring of service health (load, memory, PV/UV, error rates) and real‑time data dashboards.

3. Preventing Oversell

To avoid duplicate ticket sales, the system uses:

Distributed cache with pre‑decremented inventory.

Database unique keys on session ID and seat ID to reject duplicate inserts.

Stock expiration timestamps that release locked seats after a timeout.

User purchase limits (maximum six tickets per person).

Integration with anti‑scalper protection.

Conclusion

By designing a layered backend, isolating hotspot data, shaping traffic flow, employing multi‑level caching, and implementing rigorous stability and anti‑oversell mechanisms, Alibaba’s entertainment platform achieves the performance and reliability required for ultra‑high‑concurrency ticket sales.