ClearCausal: Cross Layer Causal Analysis for Automatic Microservice Performance Debugging

It is notoriously hard to debug performance problems in dynamic distributed systems such as containerized microservices. Developers need to spend large amounts of manual efforts to troubleshoot those performance bugs, which often incur high cost in terms of time and resources. In this paper, we present ClearCausal, a cross-layer causal analysis framework for microservices which performs fine-grained causal analysis between application function traces and infrastructure metric data to pinpoint both the root cause service and the root cause function. ClearCausal combines information theory-based causal inference anomaly detection, and dynamic dependency graph to achieve highly accurate root cause analysis. We have implemented ClearCausal and evaluated it using real performance bugs triggered in three benchmark microservices. Our results show that ClearCausal can accurately localize the buggy services and functions with much higher accuracy than other alternative solutions. Specifically, in all the tested bugs, ClearCausal successfully ranks the root cause service as the top candidate and ranks the root cause function in the top two. ClearCausal is lightweight, which makes it practical for real world production environments.