Automating End-to-End Test Suite Design for Microservices with Large Language Models

Abstract

Graphical user interface (GUI) tests are used to validate end-to-end business flows. In microservice-based systems, a single user action may trigger a chain of inter-service calls, which require verification. In practice, end-to-end test suite design is largely manual and hard to maintain. We propose a multi-agent workflow that takes as input use cases, microservice interface specification, architecture documentation, and a GUI description, and produces executable Selenium tests with test oracles. We evaluate the approach on open-source microservice systems by comparing the generated suites with manually created benchmarks. During test execution, we collect execution traces to evaluate the coverage of microservice endpoints. In addition, we evaluate the fault detection capability of the generated test suites by seeding faults into the systems under test.