AREX: Java Instrumentation‑Based Automated Regression Testing Platform for Backend Services

In complex online business systems, ensuring the correctness of modified code is challenging due to frequent iterations, high manual testing costs, and difficulty reproducing production issues; traditional automated testing often falls short in coverage and efficiency.

AREX addresses these problems by copying real production traffic into a test environment, using Java instrumentation to collect request/response data and automatically mock external interactions such as database writes, message queues, and Redis, thereby enabling high‑coverage regression testing without generating dirty data.

The core technical principle relies on the java.lang.Instrument package introduced in JDK 1.5, which allows runtime bytecode modification. AREX builds on this foundation using the ByteBuddy library to weave instrumentation code into target classes.

Key implementation details include:

Trace propagation across threads by wrapping Runnable and Callable objects with a RunnableWrapper that carries a TraceTransmitter context.

Instrumentation of ForkJoinPool tasks via a weak‑cache mechanism to preserve trace continuity without altering the original class structure.

Support for various asynchronous frameworks (Reactor, RxJava, Apache Async HTTP client, Lettuce) by instrumenting their callback mechanisms to maintain trace integrity.

Version management that detects component versions (e.g., Jedis) before class loading to apply the correct bytecode injection.

Time mocking that records timestamps during data collection and replays them by intercepting System.currentTimeMillis and System.nanoTime calls.

Local cache capture and mock for in‑memory data, ensuring consistent test results across environments.

Class‑loader isolation to keep the AREX agent separate from application code while still allowing necessary interactions.

AREX offers several advantages: low integration cost (no code intrusion), high test case stability, support for write verification, rapid local debugging of production issues, and intelligent resource management that throttles data collection under heavy load.

The project originated within Ctrip’s ticketing system, has been open‑sourced, and is now used across multiple business units for automated regression testing, inviting community contributions.