LLM-Based Robustness Testing of Microservice Applications

A recent study published on arXiv (2605.14202) explores the application of large language models (LLMs) for testing the robustness of microservice applications. The researchers implemented seven prompting strategies across three open-source LLMs (ranging from 14B to 70B parameters) focusing on two architecturally different systems: a Java monolingual system comprising six services and nine failure modes, and a polyglot system with 27 services and 14 failure modes. This resulted in 38 valid runs and 663 generated tests. A significant finding indicates that the prompt strategy accounts for more variability in test diversity than the size of the model; notably, a Structured prompt completely diminished diversity, while a single model...