AI-driven fault diagnosis from textual system logs

Kiss, Áron and Nehéz, Károly and Hornyák, Olivér (2025) AI-driven fault diagnosis from textual system logs. ANNALES MATHEMATICAE ET INFORMATICAE, 61. pp. 156-170. ISSN 1787-6117

Preview

Text 156_170_kiss.pdf - Published Version Download (690kB) | Preview

Abstract

The increasing complexity and scale of microservice-based systems pose major challenges for ensuring reliability and operational continuity. Multimodal fault diagnosis integrating logs, metrics, and traces has emerged as a key approach for improving anomaly detection, failure type identification, and root cause localization. Graph Neural Networks (GNNs) show strong potential for modeling intricate service dependencies and fault propagation patterns in such systems. This study presents a systematic review of state-of-the-art graph-based multimodal diagnostic frameworks. We compare existing methods in terms of diagnostic accuracy, scalability, computational cost, and implementation complexity, and analyze representative public datasets and benchmark systems. We highlight key challenges, including generalization, explainability, online applicability, and outline promising directions for future research. In addition, we report preliminary findings from our own experiments, which suggest that Transformer-based models provide a promising foundation for multimodal fault diagnosis in enterprise microservice systems. These early results motivate our ongoing work toward hybrid architectures that combine the strengths of Transformers and GNNs.

Actions (login required)

Edit Item