DynaCausal: Dynamic Causality-Aware Root Cause Analysis for Distributed Microservices

TL;DR

DynaCausal tackles root cause analysis in dynamic microservice ecosystems by jointly modeling time-varying system interactions and causal fault propagation. It introduces a multi-modal dynamic alignment, a system-interaction-aware representation learner (Transformer + HA-GAT), and a Causal Representation Discrimination module with Temporal Causal Disentanglement and Spatial Causal Ordering to separate true faults from normal fluctuations and to prioritize genuine root causes. The method yields robust, interpretable root-cause diagnoses and outperforms state-of-the-art baselines on RCAEval-RE2-OB and a large-scale RCA benchmark, with consistent gains across top-k metrics. This work advances RCA for cloud-native systems by addressing cascading propagation, noise resilience, and ranking bias, enabling more reliable and actionable diagnostics in highly dynamic environments.

Abstract

Cloud-native microservices enable rapid iteration and scalable deployment but also create complex, fast-evolving dependencies that challenge reliable diagnosis. Existing root cause analysis (RCA) approaches, even with multi-modal fusion of logs, traces, and metrics, remain limited in capturing dynamic behaviors and shifting service relationships. Three critical challenges persist: (i) inadequate modeling of cascading fault propagation, (ii) vulnerability to noise interference and concept drift in normal service behavior, and (iii) over-reliance on service deviation intensity that obscures true root causes. To address these challenges, we propose DynaCausal, a dynamic causality-aware framework for RCA in distributed microservice systems. DynaCausal unifies multi-modal dynamic signals to capture time-varying spatio-temporal dependencies through interaction-aware representation learning. It further introduces a dynamic contrastive mechanism to disentangle true fault indicators from contextual noise and adopts a causal-prioritized pairwise ranking objective to explicitly optimize causal attribution. Comprehensive evaluations on public benchmarks demonstrate that DynaCausal consistently surpasses state-of-the-art methods, attaining an average AC@1 of 0.63 with absolute gains from 0.25 to 0.46, and delivering both accurate and interpretable diagnoses in highly dynamic microservice environments.

Figures (4)

• Figure 1: Multi-Modal Monitoring Data and a Service Invocation Graph In a Microservice System. The Left Part Illustrates the Forms of Metrics, Logs, and Traces, Whereas The Right Part Depicts An Example of Service Invocation Graph.
• Figure 2: Network Delay Fault Case in RCAEVAL-RE2-OB: Metrics and Root Cause Scores. (a) CPU of CurrencyService Has Same Fluctuation Pattern in Abnormal/Normal Periods; Paymentservice and Adservice Memory Shows More Deviation After Checkoutservice Fault Injection. (b) Existing Methods’ Root Cause Scores for Paymentservice and Adservice Exceed That for Checkoutservice (Ground Truth).
• Figure 3: Framework of DynaCausal: Multi-Modal Data Dynamic Alignment, System Interaction-Driven Representation Learning, Causal Representation Discrimination with TCD (Temporal Causal Disentanglement) and SCO (Spatial Causal Ordering), as well as Root Cause Analysis.
• Figure 4: Parameter Sensitivity Analysis of DynaCausal on D2.