Root Cause Analysis of Anomalies Based on Graph Convolutional Neural Network

Abstract

With the gradual increase of network complexity and network scale in the cloud environment, Root Cause Analysis (RCA) of node failures has become a systematic problem of great research significance. This paper proposes Graph-Attention-Sage (GASage) algorithm, which is a fault RCA algorithm and scheme. The algorithm solves the RCA by incorporating TOPK sampling and Attention-Aggregation with GraphSage algorithm in large-scale and complex microservice network environment. The GASage algorithm is based on graph convolutional neural network and graph attention mechanism, which profoundly combines the characteristics of network fault RCA problems. TOPK sampling mechanism is applied in GASage to select the neighboring nodes with the top [Formula: see text] highest correlation as the objectives to be aggregated. GASage adopts an attention mechanism when aggregated, which aggregates the features of the central node according to the weights of the adjacent nodes and the central node. The weight aggregation method can greatly improve the node representation effect in the RCA scenario. The empirical results of our experiments have demonstrated that the model learned with GASage can outperform other model with the same learning framework and achieves more than 10% improvement in precision.