By Mohamed Sharafath M, Comcast India Engineering Center Praveen Manoharan, Comcast India Engineering Center Aravindakumar Venugopalan, Comcast India Engineering Center
In modern distributed systems, the complexity and scale of operations often lead to challenging issues in
identifying the root causes of system failures [1] ]. Traditional ways of finding out why something happened might not work well with these complicated systems, especially if they only use metrics or logs data. The huge volume of data makes manual tracing and debugging of issues impractical in a time crunch situation. The inherent limitations of isolated data sources often result in prolonged downtime, increased operational costs, and hindered system performance. Our proposed solution seeks to automate the construction of microservice dependencies by leveraging causal discovery techniques with multi-variate time-series data. With an increasing focus on explainability in many domains, causal inference has attracted much attention in the industry [2] ]. In this paper, we consider a fault in microservices as an intervention in causal inference. The Bayesian-based causal inference algorithms [3] are applied to the constructed dependency graph tree at each level. This facilitates the swift identification of the likely root cause path of microservice failures. Such prompt analysis empowers site reliability engineers (SREs) to make informed, data-driven decisions. In this paper, we discuss how implementing Causality based instant Root Cause Analysis (RCA) methods in AI for Information Technology Operations (AIOps) platforms improves reliability for efficient triaging to reduce Mean Time to Repair (MTTR).