nodeexporter container should utilize "host" networking driver to also collect host level network metrics

[Tomaz Muraus]

Right now nodeexporter and other containers as part of this deployment utilize bridge network driver.

Even though / host file system is mounted into this container, network interface metrics will only be collected for the nodeexporter container and not for host.

The reason for that is that /proc/net -> self/net is a symlink which gets resolved to self/net inside the container which means only container metrics will be retrieved.

To also collect metrics for other host interfaces, the easiest approach is to change network type to host:

  nodeexporter:
    image: prom/node-exporter:v1.3.0
    container_name: nodeexporter
    volumes:
      - /proc:/host/proc:ro
      - /sys:/host/sys:ro
...
    restart: unless-stopped
    expose:
      - 9100
    ports:
      - 127.0.0.1:9100:9100
    # We run it in host network mode so we get access to metrics for all the
    # interfaces
    network_mode: host

And then prometheus config also needs to be updated to utilize internal IP for the nodeexporter target.

For example:

scrape_configs:
  - job_name: 'nodeexporter'
    scrape_interval: 30s
    static_configs:
      #- targets: ['nodeexporter:9100']
      # If network_mode host is used for node exporter, we need to use host
      # internal IP
      - targets: ['192.168.168.10:9100']

If there are no objections, I will open a PR which explains this in the readme.

[Amir Zarrinkafsh]

That's what https://github.com/stefanprodan/dockprom/blob/master/docker-compose.exporters.yml is for.

Happy for you to PR an explanation and referring to that file in the README though.

[ZileXa]

That file isn't even mentioned in the Readme. I suppose we should use the contents of that file instead, for those 2 containers? Why not simply have 2 complete compose files?

[Amir Zarrinkafsh]

The whole point of this issue and discussion was regarding updating the README.

You can use the extend syntax, replace the networking mode in the main docker-compose.yml or override/pull in the docker-compose.exporters.yml.

There's not just a single way to achieve this outcome.

[ZileXa]

@nightah thanks I didn't know about using multiple compose files! But I will stick with one for now :)

But the exporters.yml doe not contain this part below, without this, since nodeexporter and cadvisor are not in the same docker network as Prometheus, isn't this required?

And then prometheus config also needs to be updated to utilize internal IP for the nodeexporter target.

For example:

scrape_configs:
  - job_name: 'nodeexporter'
    scrape_interval: 30s
    static_configs:
      #- targets: ['nodeexporter:9100']
      # If network_mode host is used for node exporter, we need to use host
      # internal IP
      - targets: ['192.168.168.10:9100']

If there are no objections, I will open a PR which explains this in the readme.

[Amir Zarrinkafsh]

Yes that's right if you plan to use the other Dockerfile you will need to also adjust your Prometheus configuration to point to the host to capture metrics.