Zero‑Intrusion Aspect Technology and Distributed Execution Platform in Qunit Automated Testing Framework

Qunit is an in‑house API automation testing framework built on JUnit that supports HTTP, Dubbo RPC, and Hessian interfaces, encapsulating common test functions into XML tags so testers can focus on test case design rather than implementation details.

The framework faced three typical mock‑data challenges: complex third‑party data structures, heavy reliance on mock data across many test cases, and inter‑dependent mocks causing maintenance overhead.

To solve these, Qunit integrates a zero‑intrusion aspect module built with a JavaAgent (Catcher agent). The agent modifies class bytecode at runtime, enabling dynamic recording and replay of third‑party interface data without any changes to the target application’s source code.

publicReturnType method(ParamsType[] params) throws Exception{ /* ... */ }

After instrumentation the bytecode becomes:

privateReturnType $catcher$method(ParamsType[] params) throws Exception{ /* ... */ }
publicReturnType method(ParamsType[] params) throws Exception{
    ReturnType returnObj;
    try{
        if(needMock(className, methodName, params)){
            returnObj = mock(className, methodName, params);
        } else {
            returnObj = $catcher$method(params);
        }
    } finally {
        if(needCollect(className, methodName)){
            collect(className, methodName, params, returnObj);
        }
    }
    return returnObj;
}

The Catcher agent decides at runtime whether to use mock data (via needMock) or invoke the real service, and whether to record data (via needCollect), with collection performed asynchronously to avoid performance impact.

Qunit provides two modes for mock data handling:

Recording mode – captures third‑party responses, stores them locally, and allows parameterization of fields such as respCode and respMsg.

Playback mode – injects the stored mock data during test execution, enabling fast and reliable automated tests.

Example XML configuration shows how to define a pointcut in a service and use the <catch> tag to supply mock JSON responses within a test case.

As test suites grew, execution time became a bottleneck (single‑thread mvn test runs taking up to an hour). To address this, a distributed execution platform was built on top of the internal Noah environment management system, creating multiple isolated test environments and dispatching test files to them in parallel.

The platform schedules test files based on their previous execution duration, runs them concurrently across several threads, aggregates results in real time, and presents unified reports, code coverage, and task management.

In summary, the zero‑intrusion JavaAgent technique enables seamless mock data recording/replay, while the distributed execution platform dramatically reduces overall test runtime and provides a scalable foundation for future automation frameworks within the organization.