trACE - Anomaly Correlation Engine for Tracing the Root Cause on a Cloud based Microservice Architecture

Abstract

The introduction of cloud based microservices architectures has made the process of designing applications more complex. Such designs include numerous degrees of dependencies - starting with hardware and ending with the distribution of pods, a fundamental component of a service. Though microservice based architectures function independently and provides a lot of flexibility in terms of scalability, maintenance and debugging, in case of any failure, a large number of anomalies are detected due to complex and interdependent microservices, raising alerts across numerous operational teams. Tracing down the root cause and finally closing down the anomalies via correlating them is quite challenging and time taking for the present industry ecosystem. The proposed model - trACE discusses how to correlate alerts or anomalies from all the subsystems and trace down to the true root cause in a systematic manner, thereby improving the Mean Time to Resolve (MTTR) parameter. This facilitates the effectiveness and systematic functioning of different operation teams, allowing them to respond to the anomalies faster and thus bringing up the performance and uptime of such subsystems. On experimentation, it was found that trACE achieved an average cost of (in terms of time) 1.18 seconds on prepared dataset and 4.47 seconds when applied on end-to-end real time environment. When tested on a microservice benchmark running on Amazon Web Services (AWS) with Kubernetes cluster, trACE achieved a Mean Average Precision (MAP) of 98% which is an improvement of 1% to 34% over the state of the art as well as other baseline methods.