Diagnosis of Service Failures by Probabilistic Inference with Runtime Activity Dependences
Rong Chen, Yaqing Liu, Xin Ge, Hui Li · 2016
Pinpointing sources of runtime faults, especially persistent unforeseen ones, is crucial for minimizing failure impact in the surge of service-based processes in e-commerce times. This paper presents a novel probabilistic reasoning method for diagnosing what caused the failures in service-based processes, while assuming that there is no knowledge of fault types and no formal specification of activities at design time, but activity dependence traces are available in running against test cases. Our probabilistic diagnosis is statistically significant in coping with uncertain failures arising from process executions with unknown input and output values for some activities. Experiments are carried out on various scale orchestrated web services with injected faults, and the results show that our probabilistic diagnosis statistically performs better than earlier dependency-based methods.