Customer Complaint Guided Fault Localization Based on Domain Knowledge Graph

Shuoshuo Sun, Zhihua Chai, Rui Wu, Jiawei Jin, Yonggeng Wang, Wenhao Xu, Guilin Qi

Abstract: Fault localization aims to identify where the faults occur, which is a critical task for online business systems. Currently, the work of localization is manually conducted. However, in complicated scenarios where thousands of applications are interrelated, it is difficult to quickly localize the fault even for experienced experts, which results in asset losses. The consequence urges the emergence of automatic fault localization which can assist emergency personnel. Existing automatic methods rely on learning from historical failures. However, faults rarely happen in mature systems of an enterprise, leading to the shortage of historical faulty data. To tackle this problem, we propose an Unsupervised Fault Localization (UFL) method. The proposed method utilizes customer complaints to guide localization from the perspective of semantics and leverages the domain knowledge graph to alleviate reliance on historical failures. The experimental results show that UFL outperforms existing methods for fault localization.

External IDs:dblp:conf/dasfaa/SunCWJWXQ23