MicroRCA Root Cause Localization of Performance Issues in Microservices

1. 通过调用关系构建调用图。其中还包括了服务部署于host上的边。
2. 对边进行异常检测，留下有异常的节点以及和它们相连的节点组成的子图
3. 计算边的权重和节点的分数
4. PPR

相比MonitorRank的优势是能够避免RC和frontend离得远导致的关联性不强的情况

This is because MonitorRank calculates the similarity based on the correlation between front-end services and backend services. The anomaly of microservice payment decreases the correlation during propagation and thus the root cause localization fails.

相比MicroScope的优势是能检测出来那种非末端叶节点的根因

This is because Microscope traverses the graph based on the detected anomalies and put the anomalous child nodes into a list of potential causes, which makes it fail to identify the dominating nodes when alarms are reported from child nodes.

Comments

Comments