Notifications are part of the Presentation layer as explained in Platform Overview. The notification framework runs in the Platform Stack and delivers real-time event alerts to an endpoint of your choice.
The framework has three components:
-
Views: each NVIDIA Minerva UI view (Anomaly, Trends, or Incidents) can be configured to send notifications. Views give you fine-grained control over which events trigger a notification, which dedupe key to apply, and which endpoint to send notifications to.
-
Endpoints: the destination for notifications. You can configure any number of endpoints, each with a unique name and type.
-
Templates: define the format of the notification body for a given endpoint type.
For setup instructions, see Configure Notifications.
Events
Default event configurations are provided based on learnings from existing production deployments. You can modify the defaults and configure custom event types.
Raw Events
Raw events include syslog, JSON logs, and SNMP traps. For example, you can select a specific SNMP trap type such as "coldStart", or a specific log label from a collective learning deployment such as "Parity Error" and receive notifications based on those parameters.
Generated Events
Generated events are produced by NVIDIA Minerva features. Some examples:
-
ML anomalies on metrics or logs: select which ML-based anomaly types to be notified on, for example "GPU Temperature Anomalies" or "Rare Syslogs".
-
Finite state machine (FSM) events: select BGP and interface state events such as "Link Flaps".
-
Auto-correlation events: select which raw or generated events to auto-correlate using ML, and which correlated events to be notified on.
-
Threshold events on metrics: NVIDIA recommends implementing ML-based metric anomalies to minimize false positives, but manually configured threshold notifications are also available.
Deduplication
Deduplication reduces redundant notifications. You can select any number of dimensions to form a dedupe key. Notifications with a matching dedupe key are suppressed within a configurable time window. For example, selecting both "Device Name" and "Interface Name" as a dedupe key for interface down events prevents a flapping link from flooding the notification endpoint.
Supported Endpoints
-
ServiceNow (see ServiceNow for setup details)
-
Kafka
-
Syslog
-
Email
-
Slack
-
Webhook
Last updated: