LATENCY-AWARE ROOT CAUSE INFERENCE FOR CLOUDNATIVE PAYMENT PLATFORMS USING HETEROGENEOUS GRAPH ATTENTION MECHANISMS

Cloud-native payment platforms face unprecedented challenges in maintaining ultra-low latency and high reliability while handling millions of concurrent transactions. When performance anomalies occur, identifying root causes quickly becomes critical for minimizing financial losses and preserving customer trust. This paper proposes a novel latency-aware root cause analysis framework that leverages heterogeneous graph attention networks to model the complex dependencies and diverse entity types inherent in modern payment infrastructures. Our approach constructs a comprehensive heterogeneous graph representation that captures microservices, infrastructure components, and transaction flows, while incorporating temporal latency patterns through a specialized attention mechanism. The hierarchical attention architecture employs node level and semantic-level attention to distinguish the relative importance of different failure propagation paths, enabling precise localization of latency bottlenecks and failure root causes. Experimental validation on real-world payment platform datasets demonstrates that our method achieves superior accuracy in identifying root causes compared to existing approaches, while significantly reducing mean time to resolution. The framework proves particularly effective in handling the dynamic nature of cloud-native environments where service topologies and workload patterns constantly evolve.