TopoEvo: Topology-Aware Multi-Agent Framework for Root Cause Analysis in Microservices
Researchers have introduced TopoEvo, a self-evolving multi-agent framework that incorporates topology awareness for root cause analysis (RCA) in microservices. This innovative framework tackles issues like noisy multimodal observability data (metrics, logs, traces), cascading failures, and topology drift caused by autoscaling and rolling updates. Current LLM-based RCA agents lack topology awareness and are prone to symptom-amplification bias, often misidentifying root causes. TopoEvo integrates graph representation learning with structured reasoning that respects topology constraints. It features Metric-orthogonal Multimodal Alignment (MOMA), which breaks down metric embeddings into complementary subspaces and aligns logs and traces contrastively to minimize redundancy and sparsity, resulting in consistent node representations. The research is published on arXiv with the identifier 2605.15611.
Key facts
- TopoEvo is a topology-aware self-evolving multi-agent framework for root cause analysis in microservices.
- It addresses noisy multimodal observability data, cascading failure propagation, and topology drift.
- Existing LLM-based RCA agents are topology-agnostic and suffer from symptom-amplification bias.
- TopoEvo couples graph representation learning with structured, topology-constrained reasoning.
- It introduces Metric-orthogonal Multimodal Alignment (MOMA) to reduce modality redundancy and sparsity.
- The paper is available on arXiv with identifier 2605.15611.
Entities
Institutions
- arXiv