Evolution of Qunar Flight Ticket Transaction System Architecture and Database Refactoring

Gong Ke from Qunar presents the architectural evolution of the domestic flight ticket transaction system, describing the initial simple three‑system structure with per‑agent sharding, an order index, and the challenges caused by policy imports, monolithic code, and costly deployments.

To address these issues, the team applied service‑oriented decomposition based on high QPS, business module independence, and common services, extracting policy, ticket issuance, and payment functionalities into separate services, thereby reducing coupling and deployment risk.

As the core system grew to tens of thousands of lines of code, a refactor was undertaken: a unified main order flow was defined with four abstract interfaces (order creation, payment, pre‑validation, detail retrieval), allowing sub‑systems to implement these interfaces and enabling parallel order creation.

Database challenges such as online schema changes and field explosion led to the adoption of pt‑online‑schema‑change, an order‑extension key‑value table, and eventually a non‑relational JSON‑based order detail table with sections and versioning, supporting flexible business concepts without altering the schema.

The new unified order table stores only a key_value and version, while each section holds JSON data, enabling daily sharding, simplified read/write control, and easier migration to Elasticsearch for analytics.

For monitoring, the team built an ELK stack (Flume, Kafka, Logstash, Elasticsearch, Kibana) to collect, parse, and visualize intercept logs, providing dashboards and alerting via QTalk for operational issues.

The final architecture presents a modular backend with clear service boundaries, unified interfaces, scalable storage, and robust logging, greatly improving maintainability and system resilience.