GSRFC: A Microservice Application QoS Violation Classifier Based on GraphSAGE and Random Forest

As cloud computing and containerization technologies rapidly develop, modern applications shift from monolithic to microservice architectures. Meeting service-level objectives through dynamic resource allocation is crucial for latency-sensitive apps, yet it is challenging due to microservice dependencies and dynamic API workloads. We present GSRFC, a hybrid binary classifier combining GraphSAGE and Random Forest. GSRFC predicts QoS violation probabilities under given resource quotas and API loads. It turns microservice apps into directed graphs using invocation traces, extracts global topological embeddings with multi-layer GNNs, and classifies with Random Forest. By modeling microservice interactions with graphs, GSRFC captures system states for better classification accuracy. Experimental results demonstrate that GSRFC significantly outperforms baseline models across two open-source benchmark applications of varying scales.