As cloud services shift from monolithic architectures to microservices, post-failure fault and anomaly detection becomes increasingly challenging due to cascading effects across interdependent services and the overwhelming volume of heterogeneous logs and metrics. We propose FNoDe (Faulty Node Detection), a framework that integrates application logs, performance metrics, and distributed traces into a unified graph structure to detect both the root cause and type of anomaly. By leveraging a graph convolutional network (GCN), FNoDe learns system representations under normal and anomalous states from historical microservice data and uses these embeddings to classify new system states. Evaluated on five public benchmarks and two in-house microservice systems, FNoDe outperforms traditional methods by 20–30% in accuracy and maintains competitive performance with state-of-the-art frameworks, while also offering interpretability through XAI techniques.