System Complexity Modeling and Anti‑Corruption Governance at Qunar

Author Li Jiaqi, technical director at Qunar, introduces the background of the 2022 service slimming project that reduced codebase size by 50% and the subsequent need to prevent system re‑inflation as business evolves.

To sustain the slimming results, Qunar launched a system anti‑corruption governance project aimed at managing and slowing uncontrolled system expansion, focusing on preventing ineffective code and low‑value application additions.

The work began by defining "corruption" as increased system complexity, using system complexity metrics as the basis for measurement.

Various complexity assessment models were reviewed, including The Open Group Conference 2015 model, a paper from the National University of Defense Technology, and McCabe's cyclomatic complexity, among others.

Qunar then built its own complexity model, defining dimensions such as external demands, system characteristics, integration difficulty, and impact, selecting eight key dimensions (six core, two auxiliary) after brainstorming and filtering.

Metrics are normalized to a 0‑10 scale, weighted, and include factors such as effective lines of code and code coverage.

The model treats a system as the basic unit, aggregating the complexity of its constituent applications, and distinguishes core from non‑core systems for fair comparison.

Implementation involved integrating metadata from JIRA, release systems, Sonar, tracing tools, chaos engineering configurations, and coverage systems to feed the model.

A dashboard was built to visualize system complexity, enabling queries and tracking across 1800+ applications, 28 business domains, and 120 systems.

The anti‑corruption governance mechanism sets an annual complexity growth limit of 10% (≈0.85% per month), calculates growth based on code line increase weighted by its model contribution, and automatically creates remediation tasks in JIRA when thresholds are exceeded.

Results show that overall complexity growth is controlled at 1.14%, with most governance plans completed, demonstrating the effectiveness of the model and process.

The article concludes that the systematic approach to defining, measuring, and governing system complexity can inspire similar initiatives in other organizations.