# Append this job to Prometheus scrape_configs on a GPU cluster. Service name,
# namespace and port below must match your dcgm-exporter Helm release.
job_name: dcgm
kubernetes_sd_configs:
  - role: endpointslice
    namespaces: {names: [gpu-monitoring]}
relabel_configs:
  - source_labels: [__meta_kubernetes_service_name]
    regex: dcgm-exporter
    action: keep
  - source_labels: [__meta_kubernetes_endpointslice_port_name]
    regex: metrics
    action: keep
  - source_labels: [__meta_kubernetes_endpointslice_endpoint_node_name]
    target_label: node