Practical AI with Machine Learning for Observability in Netdata

ML for observability can be challenging, given the uniqueness of each workload. However, we can leverage ML to detect individual component anomalies, even if they are sometimes noisy/imprecise. At Netdata, we use ML models to analyze the behavior of individual metrics. These models adapt to the specific characteristics of each metric, ensuring anomalies can be detected accurately.

The power of ML becomes evident when these seemingly noisy anomalies converge across various services, serving as indicators of something exceedingly unusual. ML is an advisor, training numerous independent models for each individually collected metric to achieve anomaly detection based on recent behavior.

When multiple independent metrics exhibit anomalies simultaneously, it is usually a signal that something unusual is occurring. This approach to ML can be instrumental in uncovering malicious attacks and, in many cases, predicting combined failures across seemingly unrelated components.