Extracting microservices from monolithic applications using consistent graph enhanced Graph Transformer

With the continuous development of cloud computing, the advantages of microservice architecture have become increasingly obvious compared with monolithic programs. Consequently, numerous firms are engaged in research aimed at transitioning legacy monolithic applications to microservices, thereby enabling them to maximize the advantages of cloud-based deployment. To solve the problem that manual decomposition of microservices is both time-consuming and labor-intensive, while traditional automated microservice decomposition methods cannot effectively integrate the rich structural and semantic information of single programs, we propose a microservice extraction method based on consistent graph clustering (GC-VCG). Initially, a static analysis strategy is employed to extract dependencies between classes within the monolithic application, as well as the textual information utilized in the class creation process, to construct both the structural and semantic views. Subsequently, a consistent graph enhanced Graph Transformer is utilized to learn a unified graph from both structural and semantic views. Lastly, the k-means clustering algorithm is applied to cluster the nodes, thereby identifying candidate microservices. To verify the effectiveness of GC-VCG, this paper compares it with multiple baseline methods on four publicly available monolithic applications. The results show the effectiveness of GC-VCG.