Distributed Fault Root Cause Localization and Self-Healing Strategy Generation Based on Causal Inference

Correlation-based RCA becomes susceptible to cascading failures in distributed cloud-native systems. Methods: As part of our proposal, we suggest CIRCA-SH, a causal reasoning-based closed-loop system that: (i) identifies anomalies based on metrics/tracks/logs, (ii) acquires a service-level causal structure model using intervention, (iii) orders the root causes according to their counterfactual probability and (iv) devises safe self-healing measures by evaluating counterfactuals with risk restrictions. In RCAEval data (sock store, online boutique, train tickets), CIRCA-SH performs better in Precision@3 by 7-24% over a robust baseline, decreasing the average recovery time by 18-31 times which is statistically meaningful (p < 0.01, 95 percent confidence interval). The causal explanations facilitate interpretability of localization, and possible remedial actions, which will be addressed further as we explore scalability constraints and future online causal modifications.