Networking Minerva User Guide
Networking Minerva User Guide
Breadcrumbs

Configure Enhanced Auto-Correlation

Enable the enhanced NVIDIA Minerva auto-correlation workflow during initial onboarding by importing the recommended root-cause feed views and connecting them to a correlation namespace. These feed views generate threshold events for the enhanced root-cause recipe; they are not intended primarily as display charts.

For the general namespace, endpoint, and view workflow, see Configure Auto-Correlation. For general bundle import behavior, see Import Dashboards. The Reference Configuration Package also includes detailed verification and tuning guidance alongside the enhanced auto-correlation bundle.

Before You Begin

  • Complete Import Derived Measures before importing dashboard or feed-view bundles.

  • Configure the target user tenant and sign in to that tenant as its Tenant Admin.

  • Confirm that telemetry for the intended fabric, host, GPU, and NVLink signals is being ingested.

  • Make the current Reference Configuration Package available on a system that can access the NVIDIA Minerva UI.

Import the Enhanced Feed Views

The dedicated bundle is located in the Reference Configuration Package at:

minerva-configs/dashboards/user-tenant/auto-correlation-rootcause/ac_recommended_rootcause_feed_views_filtered.json

It is also included in the user-tenant combined dashboard import set.

  1. In the target user tenant, in the upper-right corner, click Settings menu and select Actions > Import.

  2. Select the dedicated enhanced auto-correlation bundle, or use the user-tenant combined set if the other initial dashboards have not yet been imported.

  3. Click Import and confirm that the grouping dashboard and all ac_feed_* Trends views import without errors.

Create the Namespace and Endpoint

  1. Go to settings menu > Settings > System Configuration. Next to Enable Auto-Correlation, click +.

  2. In Auto-Correlation Namespace, enter recommended_rootcause. NVIDIA Minerva adds a prefix based on your tenant and username, so the saved name is <tenant>_<username>_recommended_rootcause.

  3. In Correlation Method, select Topology.

  4. In View Type, select Trends Page. In View Name, select the imported ac_feed_* views.

  5. Save the namespace.

  6. In settings menu > Settings > Notifications, create the correlation endpoint described in Configure Auto-Correlation:

    • Endpoint name: correlation-endpoint

    • Plugin type: syslog

    • Transport: udp

    • Server: 127.0.0.1:59000

    • Select Inherit from event's facility, Inherit from event's notificationSeverity, and Inherit from event's application.

Verify the Feed Views

For each feed that will be active during initial onboarding:

  1. Confirm Auto-Correlation is enabled and uses the namespace you created, which ends in recommended_rootcause.

  2. Confirm Notifications is enabled and selects correlation-endpoint.

  3. Confirm that no notification deduplication dimensions are selected.

  4. Keep the imported data source, event-source filters, table visualization, polling window, thresholds, and auto-correlation settings unchanged unless the site-specific tuning guidance in the Reference Configuration Package requires a change.

Nine feeds are expected to be active after import. The two CNP feeds, ac_feed_cnp_sent and ac_feed_cnp_handled, ship with notifications disabled because a zero threshold can create excessive events on fabrics with a non-zero CNP baseline. Leave these two feeds disabled during initial onboarding. Measure the site's normal CNP rate, set a threshold above that baseline, and then enable them.

Validate Onboarding

  1. Generate or wait for a signal that crosses one of the enabled feed thresholds.

  2. Open Incidents.

  3. Confirm that related events are grouped into an incident.

  4. Confirm that the incident's Root Cause field contains a classified value or Undetermined, rather than remaining empty.

  5. If incidents are missing, use the verification procedure supplied alongside the bundle in the Reference Configuration Package to confirm which ac_feed_* views the notifier loaded.

  6. If the Root Cause field is empty, confirm the namespace, view assignments, notification endpoint, and that the deployed NVIDIA Minerva build contains the enhanced root-cause recipe.

Last updated: