How Traffic Recording & Replay Boosts Test Efficiency at China Agricultural Bank

Introduction

With the continuous evolution of the Agricultural Bank of China's technical architecture and business scenarios, testing faces problems such as complex data construction, high case‑maintenance cost, and uneven quality. Four main challenges are identified: extensive regression testing after system upgrades, time‑consuming regression for complex business logic, high manual effort for test case creation and maintenance, and the need for rapid issue reproduction in production anomalies.

Traffic Recording & Replay Overview

Traffic recording and replay technology quickly captures interface traffic generated during application execution according to filtering rules. Recorded data includes request and response messages, which can be replayed in designated environments for functional regression testing or version verification, simplifying test case writing, improving efficiency, and enhancing scenario coverage.

Classification of Recording Techniques

Web‑server based recording : Captures requests by customizing the OS and web server; supports diverse request types but requires high development and maintenance cost.

Network‑stack based recording : Listens on network ports and copies packets; minimally impacts the application but may affect network performance.

Application‑level recording : Uses AOP to intercept requests; non‑intrusive to code and requires no application modification, though it consumes some server resources.

Design Solution

In July 2022, the Tianjin R&D testing team of the Agricultural Bank established a special project to build a traffic recording and replay tool using open‑source and self‑developed components. The platform provides configuration management, node monitoring, traffic management, replay execution, batch execution, and result analysis.

Architecture

The tool follows a front‑back separation architecture. A probe installed on the application records traffic, while a proxy module listens, processes replay requests, and forwards information to a management module that handles storage and configuration. Users interact via a unified web UI to create configurations, record traffic, and query replay results.

Data Layer Strategy

To support fast queries on massive recorded traffic, the tool stores online traffic and replay results in Elasticsearch, achieving millisecond‑level response for millions of records. Smaller management data resides in MySQL, and Redis queues handle ordered batch replay.

Batch Replay & Result Query

Recorded traffic can be manually filtered into batches and replayed in the original call order to simulate mixed transaction scenarios. The batch execution provides result statistics and success‑rate trends, helping testers monitor interface performance and adjust cases promptly.

Message Comparison Checkpoints

After replay, the tool compares recorded response messages with current responses, ignoring noise fields (e.g., timestamps, random numbers) by configuring checkpoints such as response codes. Matching checkpoints indicate successful replay.

Header Replacement

To handle session expiration during web‑application replay, the tool can replace header parameters in real time, ensuring continuous replay of recorded traffic.

Case Export

Recorded traffic cases can be exported as Collection V2 JSON files, compatible with Postman, Postwoman, and other API testing tools, enabling reuse without manual message assembly.

Lifecycle Management

Users can configure traffic data lifecycle policies per application and environment. Unused traffic not added to batches is periodically cleaned, while batch‑included traffic can be retained permanently for storage or replay.

Application Effect

The platform has been gradually rolled out within the Tianjin R&D department, with over 20 modules integrated (67% web, 13% online, 20% mobile). A total of 4.06 million interfaces have been recorded, and batch replay success rates exceed 85% after process optimization.

Conclusion

Traffic recording and replay provides a novel approach to automated interface testing, applicable to version verification and regression testing. By converting live traffic into test cases, it dramatically reduces manual effort in data preparation and script writing, offers more realistic coverage of business scenarios, and enhances defect detection, thereby improving testing efficiency and lowering costs.