Efficient Fault Localization in a Cloud Stack Using End-to-End Application Service Topology

2025-09-05

AI Summary

Facing the dual challenges of explosive observability data and inefficient root cause localization in cloud-native microservices, this paper proposes TA-RCD, a topology-aware root cause detection algorithm. TA-RCD operates on end-to-end service topologies and jointly models call-chain structures with multi-source metrics. It incorporates topology-guided metric filtering and causal inference to enable interpretable identification of anomaly propagation paths. Evaluated on real-world microservice deployments and systematic fault-injection experiments, TA-RCD achieves over a two-fold average improvement in Top-3 and Top-5 recall compared to state-of-the-art RCD methods. It significantly enhances localization accuracy, efficiency, and interpretability. This work establishes a lightweight, topology-driven paradigm for root cause analysis, advancing elastic assurance for cloud services.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning

Application Category

• Responsible Web: Algorithmic accountability and transparency on the web
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Cloud application services are distributed in nature and have components across the stack working together to deliver the experience to end users. The wide adoption of microservice architecture exacerbates failure management due to increased service components. To be effective, the strategies to enhance the application service resilience need to be autonomous and developed at the service's granularity, considering its end-to-end components. However, the massive amount of observability data generated by all these components across the service stack poses a significant challenge in reacting to anomalies and restoring the service quality in real time. Identifying the most informative observability data from across the cloud service stack and timely localization of root causes of anomalies thus becomes crucial to ensure service resilience. This article presents a novel approach that considers the application service topology to select the most informative metrics across the cloud stack to support efficient, explainable, and accurate root cause identifications in case of performance anomalies. The usefulness of the selected metrics is then evaluated using the state-of-the-art Root Cause Detection (RCD) algorithm for localizing the root cause of performance anomalies. As a step towards improving the accuracy and efficiency of RCD, this article then proposes the Topology-Aware-RCD (TA-RCD) that incorporates the end-to-end application service topology in RCD. The evaluation of the failure injection studies shows that the proposed approach performs at least 2X times better on average than the state-of-the-art RCD algorithm regarding Top-3 and Top-5 recall.

Problem

• Localizing root causes of performance anomalies in cloud services
• Selecting informative metrics across cloud stack for fault diagnosis
• Improving accuracy and efficiency of root cause detection algorithms

Innovation

• Leveraging end-to-end service topology for metric selection
• Developing Topology-Aware-RCD algorithm for improved accuracy
• Enhancing root cause detection with explainable anomaly identification