Traffic Replication and Replay Platform for JD APP: Design, Features, and Operational Impact

Software testing is a crucial part of ensuring the quality and smooth operation of the JD APP shopping process, especially during high‑traffic events like 618 and 11.11, where large‑scale load testing is required.

To accurately simulate user scenarios for pressure testing, JD captures real user search terms from front‑end input logs, stores them, and replays them during tests, but manual processing of this data is costly.

To address this, JD developed a traffic‑replication and replay solution, defining traffic as the data transmitted during user visits, which can be copied and either stored for offline use or forwarded to a new server for online testing.

The Sniper Test Platform was built to provide an easy‑to‑use, HTTP‑only traffic replication and replay service with features such as diverse functionality, convenience, data security, traceability, and scalability. It consists of five core modules: traffic control and management center, traffic replication service, traffic processing service, traffic replay service, and traffic tracking & persistence center.

Key platform capabilities include:

Copying traffic from specific services identified by domain names.

Processing copied traffic to filter out write‑heavy interfaces and avoid dirty data.

Replaying processed traffic in parallel to generate realistic load.

Persisting test results for analysis and trend tracking.

Normalised load testing is performed periodically, following five steps: environment detection, traffic capture, warm‑up testing, formal testing, and result analysis, enabling continuous performance monitoring without manual script creation.

Dynamic regression testing leverages traffic replication and replay to compare responses of new service versions against live traffic, improving release success rates.

Operational results show that the platform has tracked around 30 core services, providing valuable capacity evaluation data, though current limitations include support only for HTTP protocols, periodic (not event‑driven) testing, and challenges with replaying write interfaces.

Future work aims to extend protocol support, integrate with continuous delivery pipelines for event‑triggered testing, and improve handling of write‑heavy interfaces.

Overall, the Sniper Test Platform automates traffic‑based load and regression testing, reducing manual effort and enhancing product quality, while acknowledging that testing techniques still have much room for advancement.