Intelligent profile-driven drift detection

Abstract

Techniques are disclosed for detecting a drift experienced by computing system(s). The system generates multiple snapshots as part of a drift detection process. Each snapshot contains state information of a computing system. Based on the snapshots, the system generates metrics sets according to a general specification. The general specification defines metrics generally suitable for detecting drift in the computing system(s). Based on the circumstances of the drift detection process, the system generates a custom specification. The system optionally employs trained machine learning model(s) for custom specification generation. The custom specification defines modifications to the metric sets designed to make the metric sets more suitable for the circumstances of the drift detection process. The system modifies the metric sets according to the custom specification. Subsequently, the system generates flattened vectors based on the modified metric sets, and the system performs a cluster analysis on the flattened vectors to detect any drift.