Fault Diagnosis for Open Source Software Based on Dynamic Tracking
Xuliang Chen, Jianhui Jiang, Wei Zhang, Xuzei Xia · 2020
As the complexity of software increases, the fault diagnosis of software encounters huge challenges. The software log contains status information at runtime, so it is widely used in software fault diagnosis. But for open source software, there are problems such as lack of log, irregular log and incomplete information. To address these problems, a new fault diagnosis method based on dynamic tracing is proposed to generate open source software enhanced log. Besides, a critical function mining algorithm is proposed to select the critical functions. And the function call relationship graphs (FCRGs) can be obtained by the dynamic tracking of these critical functions during the software operating. Then, the FCRGs are detected by the graph attention network (GAT) model. For abnormal ones, a top K closest fault diagnosing algorithm are proposed to determine the root cause of failure. The experimental results on open source software Curl, Redis and Nginx show that our method is indeed rational and effective.