MBST: a bad smells testbed for microservice systems

Poor design in microservice architectures can lead to microservice bad smells (MBSs), negatively impacting system performance, reliability, and scalability. These latent MBSs, if not promptly detected and eliminated, can accumulate and cause significant damage. Since MBSs manifest differently across microservice systems and require diverse detection techniques, achieving uniform and effective detection remains a significant challenge. To address this problem, we propose a centralised testbed for identifying microservice bad smells. First, we provide a catalogue of 22 microservice bad smells that can be reasonably detected by rule-based methods, which is fully oriented towards the detection perspective and classified according to impact layer, generation phase and the best applicable detection method. Second, we propose a microservice bad smells testbed (MBST) that detects MBSs in two phases: static and runtime. This testbed achieves comprehensive and unified detection of 22 MBSs through various detection methods. Third, we propose microservice software health assessment models for timely quantification of the impact of MBSs on the system. Through experimental validation, we demonstrate the effectiveness of different smell detection methods and the accuracy of the assessment models. Compared to other open-source tools, the MBST exhibits superior accuracy, efficiency, and specificity in smell detection. Furthermore, we analyze the temporal characteristics and impact of smell occurrences under varying load conditions and detection intervals. The MBST enables continuous detection and localisation of MBSs, while assessing the system’s health and runtime performance.