An Interception Method for Non-Deterministic Noise in Traffic Replay Scenarios

This paper examines the use of real traffic recording and playback technology for multi-version testing, focusing on the importance of noise identification and processing in result comparison. It introduces a new method to distinguish between random noise and divergent behavior. Both domestic teams (e.g., Alibaba, NetEase, ByteDance) and international teams (e.g., Twitter's Diffy tool) have advanced traffic processing technologies, but challenges remain in handling non-deterministic factors. The paper presents an interception technique at the Java Virtual Machine (JVM) level, based on bytecode enhancement, which effectively controls noise. Testing shows that this technique significantly improves problem detection in core system testing, increasing defect detection by 2%. It provides an effective noise control strategy for fields requiring precise traffic replay, such as finance.