A root cause analysis framework termed InterfaceRCA achieved an average Hit Rate at 1 of 92.57 percent and an average Normalized Discounted Cumulative Gain at 1 of 95.51 percent across public benchmarks and an industrial production dataset. The findings were reported in the journal Cluster Computing.
Modern microservice architectures rely on intricate inter-component communication patterns that create significant challenges for automated root cause analysis. According to the authors, prior diagnostic methodologies exhibit two primary shortcomings. First, existing tools operate at a coarse granularity that fails to capture the complex, fine-grained characteristics of microservice application interface calls. Second, conventional approaches neglect how downstream diagnostic tasks are influenced by the feature optimization and feature fusion of multimodal operational data.
To overcome these challenges, InterfaceRCA establishes an analytical pipeline that leverages multimodal telemetry collected across microservice interfaces. The methodology starts by segregating heterogeneous data sources—specifically including system logs, numerical metrics, and distributed execution traces—based on the topology of the system call graph. Once segregated along the call graph structure, these diverse inputs undergo feature optimization and fusion through dedicated neural networks and embedding techniques.
At the center of the fault localization procedure, the architecture incorporates a Graph Attention Network based autoencoder. This Graph Attention Network autoencoder evaluates the fused interface representations and calculates individual reconstruction scores for each node within the interface graph. By computing reconstruction errors at this topological resolution, InterfaceRCA identifies anomalies and performs root cause analysis directly at the granularity of individual microservice interfaces rather than generic component boundaries.
The diagnostic capability was validated through extensive evaluations conducted on both public benchmark datasets and a real-world production dataset. InterfaceRCA demonstrated superior precision over established single-modal and multimodal baseline systems. When compared against FaaSRCA, the strongest multimodal baseline examined in the study, InterfaceRCA achieved an absolute advantage of approximately 2.93 percentage points in HR@1. Furthermore, the approach recorded an average improvement of 82.90 percentage points over MonitorRank, the weakest single-modal baseline evaluated across both benchmark environments.
Subject of Research: Root cause analysis in microservice systems using interface multimodal data
Article Title: InterfaceRCA: fault root cause analysis of microservice system based on interface multimodal data
Article References: Wang, Y., Hu, Z., Huang, H., Lin, W., & Li, K. (2027). InterfaceRCA: fault root cause analysis of microservice system based on interface multimodal data. Cluster Computing, 30(1), Article 16. https://doi.org/10.1007/s10586-026-06819-x
Image Credits: AI Generated
DOI: 10.1007/s10586-026-06819-x
Keywords: InterfaceRCA, root cause analysis, microservice systems, multimodal data, Graph Attention Network, autoencoder, system call graph, call graph topology, system logs, system metrics, distributed traces, feature fusion, embedding techniques, fault diagnosis, HR@1, NDCG@1, FaaSRCA, MonitorRank
News Source: Denise Maddox. (November 16, 2026). Interface Multimodal Analysis Accurately Identifies Fault Root Causes in Microservice Systems. Scienmag.



