Dynamic- X-Y: A Tool for Learning Dynamic Alert Suppression Policies in AIOps

Abstract

Although Cloud Native Network functions (CNFs) provide greater agility, manageability, and significantly lower operational costs, the reliability and performance assurance is getting increasingly complex, therefore observability tools are needed to monitor and detect anomalous events, triggering alert notifications and creation of incidents. However, most of these notifications turn out to be false alarms, leading to alert fatigue, inefficiencies, and the risk of missing critical alerts. Existing approaches for reducing alert noise rely on static policies that can quickly become outdated in dynamic IT environments. We demonstrate a novel unsupervised approach, Dynamic-X-Y, which learns dynamic alert suppression policies from historical alert data and applies them to incoming events/alerts at runtime, thereby reducing unnecessary alert notifications. Our approach achieves an accuracy of 93.93% in identifying correct alerts, outperforming the baselines by a significant margin. Additionally, we present a case study demonstrating the effectiveness of our approach vis-a-vis the No-Sunnression annroach.