digitalocean/do-dcgm-exporter

digitalocean/do-dcgm-exporter

Description: The DigitalOcean DCGM Exporter is a thin wrapper around the dcgm-exporter for DigitalOcean droplets.

Language: Go

License: Apache-2.0

Stars: 1

Forks: 0

Open issues: 2

Created: 2024-11-19T14:35:09Z

Pushed: 2026-03-19T00:15:04Z

Default branch: main

Fork: no

Archived: no

README:

DigitalOcean DCGM-Exporter

Overview

The DigitalOcean DCGM-Exporter is a thin wrapper around the DCGM-Exporter.

The following functionality is added

• configuration of a default set of DCGM fields to be monitored. Additional fields can be configured using the --collectors flag (like with dcgm-exporter).
• forwarding of the collected metrics to the in-droplet accessible DigitalOcean endpoint with static ip 169.254.169.254.
• requirement of a standalone DCGM installation with nv-hostengine serving on localhost:5555. This is to avoid conflicts with existing dcgm-exporter installations.

Exposes a /metrics endpoint serving the collected Prometheus metrics on port 9401.


Run Requirements

Requires DCGM and NVIDIA drivers to be installed.

Installation

Please see the [installation documentation](docs/install.md).

To build the DigitalOcean DCGM-Exporter manually, please see [here](docs/build.md).

Conflict with existing DCGM installation

Please note that there can only be one DCGM installation on a host. This includes an embedded DCGM process started by the NVIDIA dcgm-exporter.

Hence, to run the DigitalOcean dcgm-exporter next to the NVIDIA dcgm-exporter,

• please create a standalone installation of DCGM.
• configure the NVIDIA dcgm-exporter to connect to the remote nv-hostengine serving on localhost:5555 (via flag -r localhost:5555).

The DigitalOcean DCGM-Exporter connects to a nv-hostengine process serving on localhost:5555.

Testing Restrictions

The DigitalOcean DCGM-Exporter is a thin wrapper around the DCGM-Exporter. While this has the benefit of being able to reuse functionality, it restricts the DigitalOcean DCGM-Exporter to the boundaries setup by the DCGM-Exporter code. Specifically, variables required for mocking hardware (GPUs, NVSwitches, ...) are not exported.

As a result, this project does not contain test cases covering dcgm-exporter functionality that requires real hardware.