NRCAC: Non-Intrusive Microservice Root Cause Analysis Framework for Cloud Providers

In modern containerized cloud environments, even minor performance issues in microservice systems can significantly impact overall functionality, highlighting the critical need for cloud providers to implement efficient Root Cause Analysis (RCA) to meet SLA requirements. Existing RCA frameworks often rely on intrusive data collection techniques and algorithms requiring extensive computation, which are not only privacy-invasive but also induce considerable performance overhead and predominantly focus on tenant-specific issues, making them less suitable for cloud providers. This paper investigates whether it is possible for cloud providers to perform RCA focusing on eliminating intrusiveness and reducing overhead. To this end, we propose NRCAC, a novel RCA framework designed specifically for cloud providers. NRCAC includes eCollection, a non-intrusive data collection method utilizing eBPF, allowing cloud providers to gather data from the host kernel efficiently without directly accessing microservice applications. Additionally, NRCAC features RCD-DK, a high-performance root cause analysis algorithm which reduces the search space of causal graphs by incorporating domain-specific insights. Evaluation with both synthetic data and real-world microservice systems demonstrates the effectiveness of NRCAC in accurately identifying the root causes of anomalies, outperforming existing frameworks by 46% to 172%.