KubeCon + CloudNativeCon North America 2024

In response to challenges of limited trace capture in dynamic API tracing systems, the solution leverages Machine Learning and Cognitive approach for unbiased trace collection. Unlike existing implementations with a skewed distribution(~5%) towards normal traces, our self-adaptive system dynamically learns to prioritise and capture diverse traces, crucial for effective diagnosis of API failures and performance issues. This innovative approach significantly enhances the SREs ability to triage complex issues, leading to a game-changing reduction in Mean Time to Resolve (MTTR). The Adaptive Sampling approach analyses existing system traces and autonomously adjusts the sampling rate, eliminating manual configs. This ML-based solution outcome includes streamlined trace metric analysis, enhanced reliability work efficiency, and considerable infrastructure cost reduction through targeted trace collection, ultimately making a significant impact on operational effectiveness & reliability