Machine Learning Enrichment

Adaptive Thresholds

Distributed Kubernetes systems are anything but static. They are constantly changing and adapting to your customers and changing situations within the cluster. Static Thresholds are no longer a reliable tool for helping operators when the system needs attention. FreshTracks learns the behavior of your system over time and then calculates expectations for the metrics in your system. These thresholds are useful tools for developers and operators to identify and diagnose aberrant issues within the system.

Anomaly Detection

Humans are inherently good at pattern matching, but so are computers. At FreshTracks, we would rather have machines detect anomalies so that engineers can spend their time tackling more urgent and important problems. FreshTracks uses various statistical and machine learning methods to identify anomalies and emit them in a reliable stream. The stream is stored alongside your existing series in Prometheus to allow your team to use the anomalies to improve Prometheus alerts or highlight issues visually within Grafana.

Outlier Detection

FreshTracks can tell you why a container is performing badly compared to the others. Our anomaly detection not only finds individual metrics that stray from the expected behavior, we also find single metrics that are performing significantly different from similar metrics. FreshTracks uses dynamic time warping to identify and isolate outlier series and highlight them visually.

System Correlation

A complex web of interactions occurs between infrastructure, Kubernetes, containers and applications, and between applications themselves. FreshTracks draws from quantum mechanical techniques to understand complex interactions within your distributed system and highlight them to operators and developers as issues arise.

Data Enrichment

The calculations that FreshTracks performs on your metrics are written back to your Prometheus data source as a set of Prometheus series. This data includes standard and custom aggregations, machine learning derived dynamic thresholds and anomalies. This allows you to have complete control over the visualization using custom Grafana dashboards and alerts using the Prometheus Alert Manager.