! DEPRECATED !

This code is in the process of migrating into the RHOSO EDPM Ansible repo. Currently it exists on a fork for testing pending preparations for downstream delivery, here: https://github.com/rhos-vaf/edpm-ansible/tree/rhos-vaf/roles/edpm_telemetry_accel_exporters

This repo should be ready to be archived.

! DEPRECATED !

NVIDIA DCGM Exporter Ansible Role

This Ansible role installs and configures the NVIDIA DCGM Exporter for GPU monitoring and metrics collection.

Requirements

• Ansible 2.14 or higher
• Target systems must have:
  • Podman installed
  • Network access to pull container images

Role Variables

Variable	Default	Description
dcgm_exporter_nvidia_repo_url	"https://developer.download.nvidia.com/compute/cuda/repos/rhel9"	Base URL for NVIDIA repo
dcgm_exporter_image_tag	"4.3.1-4.4.0-ubi9"	Container image tag for the DCGM Exporter
dcgm_exporter_libnvidia_container_toolkit_version_release	"1.17.8-1"	Version of the NVIDIA Container Tools RPM to install
dcgm_exporter_nvidia_driver_module_version	"open-dkms"	NVIDIA Driver version to install if not already present

Dependencies

None.

Example Playbook

---
- hosts: gpu_servers
  roles:
    - nvidia-dcgm-exporter

With custom image tag:

---
- hosts: gpu_servers
  vars:
    dcgm_exporter_image_tag: "4.3.1-4.4.0-ubi9"
  roles:
    - nvidia-dcgm-exporter

What This Role Does

1. Installs NVIDIA Driver: Installs nvidia-driver from the NVIDIA repository if not already installed
2. Installs NVIDIA Management Library: Installs the libnvidia-ml RPM that matches the installed nvidia-driver version
3. Installs NVIDIA Container Tools: Installs the libnvidia-container-tools RPM for container GPU support
4. Updates CDI Configuration: Adds the required mount configuration to /etc/cdi/nvidia.yaml for nvidia-ml library access
5. Creates Systemd Service: Creates a systemd service file that manages the DCGM Exporter container
6. Manages Service: Enables and starts the DCGM Exporter service

Container Configuration

The role runs the DCGM Exporter in a container with the following configuration:

• Image: nvcr.io/nvidia/k8s/dcgm-exporter:{{ dcgm_exporter_image_tag }}
• Port: 9400 (Prometheus metrics endpoint)
• Devices: All NVIDIA GPUs (nvidia.com/gpu=all)
• Capabilities: SYS_ADMIN
• Environment: DCGM_EXPORTER_DEBUG=true

Service Management

The role creates a systemd service that:

• Automatically starts on boot
• Restarts on failure
• Uses podman to manage the container lifecycle

Metrics Endpoint

Once deployed, metrics will be available at http://your-server:9400/metrics