Predictive SLO Breach Prevention Using Time-Series and Graph Neural Networks

Abstract

This paper presents a novel approach for proactively preventing Service Level Objective (SLO) breaches in microservice-based architectures using a hybrid model of time-series forecasting and Graph Neural Networks (GNNs). Leveraging the temporal and topological characteristics of cloud-native environments, we propose a predictive pipeline capable of modelling dynamic service dependencies and pre-emptively flagging potential SLO violations. Experimental evaluations on simulated and real-world datasets show that our framework significantly outperforms baseline models in predictive accuracy and resource efficiency.