The Incidents page is the primary interface for high-level cluster monitoring. While Anomaly and Trends views show individual data points, Incidents aggregate related events into groups, reducing notification noise and providing a single timeline for root-cause analysis.
Incidents Overview
The Incidents view organizes correlated events into a summary table, giving you a high-level view of cluster outages and performance degradations.
|
Field |
Description |
|---|---|
|
Timestamp |
The time the correlation engine identified and grouped the related events |
|
ID |
A unique identifier for the incident, typically prefixed with the correlation namespace |
|
Devices Correlated |
All hostnames or switches involved in the incident, for example |
|
Event Count |
Total number of individual anomalies or log messages rolled up into the incident |
|
Potential Impact |
An automated assessment of the issue's severity and scope |
|
Root Cause |
The primary trigger that links the events together |
Understanding Correlation
Incidents are generated based on two primary logic types defined in the system. Knowing which type you are viewing helps in interpreting the root cause:
Dimensional Correlation
Dimensional correlation groups events based on shared metadata. For example, a rack correlation incident groups all anomalies occurring within the same physical rack. This is effective for identifying localized hardware issues, power failures, or top-of-rack switch misconfigurations.
Topology-Aware Correlation
Topology correlation groups events based on their relationship in the network fabric. For example, if an upstream spine switch experiences an anomaly and the downstream leaf switches show a spike in pause frames, NVIDIA Minerva correlates these as a single incident to show how the issue propagates.
Incident Detail and Topology Events
Selecting an incident ID opens a detailed drill-down view with a Topology Events table. The table lists every individual event that contributed to the incident.
-
Timeline View: groups key events into a single chronological timeline.
-
Granular Analysis: shows specific error types such as
HostGpuDriverVersionMismatchorLinkDownedCounteracross the correlated devices.
Last updated: