Design and Implementation of Ctrip International Ticketing Data Middle Platform

Background and challenges : Rapid growth of Ctrip's international ticket business increased data variety and complexity, raising requirements for data accuracy and real‑time availability. Producers face long update cycles, high maintenance cost, low access efficiency, and difficult roll‑backs, while consumers suffer from error propagation, slow issue detection, and environment‑related performance problems.

Principles and goals : The platform aims to integrate data resources, eliminate data silos, improve processing efficiency, and ensure quality. Core objectives include data consistency, timeliness, system robustness, traceability, rollback capability, comprehensive monitoring, cost reduction, simplified consumption, unified data model, environment problem solving, and performance enhancement.

Key technical practices :

1. Data consistency – version control using BLOB files, BitTorrent‑based P2P distribution, single‑point data generation, and combined data packaging to guarantee consistent snapshots across consumers.

2. Data timeliness – push‑pull mechanism with configurable pull‑tasks and cloud migration of Blob files to hybrid‑cloud regions, achieving >98% cost reduction and seamless cloud transition.

3. System robustness – data validation with alerts via TripPal, email, SMS; AI‑assisted anomaly detection; rollback functionality that restores historical versions without database changes.

4. Consumption‑process optimization – unified data model, automated build/deploy of model jars, and simplified access via generated client APIs.

5. Unified data governance – searchable portal for data/models, streamlined onboarding, and >90% improvement in data‑source integration efficiency.

Technical architecture overview : The platform consists of six core modules – DataSource, BlobGenerator, BlobService, DataClient, DataQuery, and Dispatcher – collaborating to ensure end‑to‑end data flow, high scalability, and maintainability.

Results : End‑to‑end latency reduced to 23 seconds (5 seconds for small datasets), server cost cut >95%, maintenance cost down 66%, data‑source onboarding speed up 90%, and >98% reduction in ineffective scheduling and GC frequency.

Future plans : Further automation (including LLM‑driven validation), stability enhancements, robustness, timeliness, and visualization of data pipelines and portal UI.

// Import client
@DataResource
private CityClient cityClient;
// Full‑list query
List<City> list = cityClient.queryList();
// Conditional query
List<City> list = cityClient.queryList(cityCode);