Web service fault diagnosis based on graph convolutional networks
zhongxuan yang, Zhichun Jia, Wanying Cheng, Xinyuan Cui, Jiatong Li, Jianyu Qi, Xing Xing · 2024
Web service composition refers to combining multiple independent web services to form a new, more complex application system. These web services can come from the same provider or different providers, and can also come from different application domains. Consequently, errors are more likely to occur during the execution of web service compositions. Although XML-type log files are generated to record errors that occur during runtime, the location of the exception may not necessarily be the service where the error occurred. Finding errors in complex service compositions is crucial, which is why we have chosen graph neural networks as a fault diagnosis method for analyzing the relationships between nodes. We analyze all possible problematic services in the entire service flow based on collected data from faulty web service compositions. First, we label known errors in a faulty service composition. Then, we design the graph structure of the service flow based on the call relationship of the service composition, i.e., constructing a topology graph with the services included in the service composition as nodes and the call relationship as edges. Using collected data from faulty service compositions, we construct features for each service, i.e., point features. Finally, we train them using GCN and perform clustering. Based on the classification results, we can identify all faults in the service flow. Through experimental comparison, we demonstrate that GCN is more effective in diagnosing faults in web service compositions