Quickstart

Get the gNMI exporter running in your cluster and collecting network telemetry in minutes.

Installation

Install the monitoring operator using the Helm chart published as an OCI artifact.

Available versions can be found at the GitHub Container Registry.

helm install gnmi-exporter \
  oci://ghcr.io/cobaltcore-dev/charts/gnmi-exporter \
  --version 0.0.0-36c6d53 \
  --namespace monitoring-system \
  --create-namespace

Creating a DeviceMonitor

A DeviceMonitor selects network devices by label and configures gNMI subscriptions for telemetry collection:

apiVersion: monitoring.wire.cobaltcore.dev/v1alpha1
kind: DeviceMonitor
metadata:
  name: spine-monitor
spec:
  replicas: 2
  deviceSelector:
    matchLabels:
      wire.cobaltcore.dev/device-role: evpn-spine
  template:
    spec:
      nodeSelector:
        topology.kubernetes.io/zone: eu-de-1a
  subscriptions:
    - name: interface-counters
      paths:
        - /interfaces/interface/state/counters
      mode: Stream
      streamMode: Sample
      sampleInterval: 10s
  encoding: json_ietf

Prometheus Integration

The operator always creates a headless metrics Service exposing port 9804. You can integrate with Prometheus in two ways.

Option A: Use an existing ServiceMonitor

If you already have a ServiceMonitor (or a Prometheus scrape configuration) that discovers Services by label, add custom labels to the metrics Service using spec.metrics.additionalLabels:

spec:
  metrics:
    additionalLabels:
      monitoring.example.com/scrape: "true"

The operator merges these labels onto the created Service. Your existing ServiceMonitor can then select it:

apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: networking-services
spec:
  selector:
    matchLabels:
      monitoring.example.com/scrape: "true"
  endpoints:
    - port: http
      path: /metrics

Option B: Let the operator create a ServiceMonitor

If you want the operator to manage its own ServiceMonitor, set spec.metrics.serviceMonitor:

spec:
  metrics:
    serviceMonitor:
      additionalLabels:
        prometheus: infrastructure
      interval: "30s"
      scrapeTimeout: "10s"

The operator creates a ServiceMonitor targeting the metrics Service. Use additionalLabels to match the serviceMonitorSelector on your Prometheus CR so it gets picked up automatically.

No monitoring resources (default)

When spec.metrics is omitted entirely, the operator does not create a ServiceMonitor. The metrics Service still exists and can be scraped manually or via pod annotations.