An Approach of Automated Anomalous Microservice Ranking in Cloud-Native Environments

Abstract

In recent years, more and more developers have been building applications based on the cloud-native architecture. Container and microservice are two essential components in the cloud-native architecture. Container technologies like Docker and Kubernetes can help developers achieve a consistent and scalable delivery for complex software applications. On the other hand, microservice technologies can facilitate the division of complex applications into multiple functionality-independent and composable components, which further increases the flexibility of applications. With the support of cloud computing platforms, cloud-native applications will be easier to manage and maintain, together with higher scalability. However, it is challenging to identify performance issues on microservices due to the complex runtime environments and the numerous monitoring metrics. Towards this issue, this paper proposes a novel root cause analysis approach. Our approach firstly constructs a service dependency graph based on the metrics collected in real time. Next, the anomaly weight of each microservice is automatically updated by extending the mRank algorithm. Finally, a PageRank-based random walk is adopted to rank root causes further, i.e. to rank potential problematic services. Experiments conducted on Kubernetes clusters show that the proposed approach achieves a good analysis result, which outperforms several baseline methods.