Add plugin for Docker

[oli-g]

Add plugin for Docker

[oli-g]

@anryko Hey, I would suggest you to run our collectd Docker image with docker-collectd-plugin preinstalled. Just run something like:

docker run \
    -d \
    --name collectd \
    --hostname foo \
    --privileged \
    --volume /proc:/mnt/proc:ro \
    --volume /etc/collectd:/etc/collectd \
    --volume /var/run/docker.sock:/var/run/docker.sock \
    chimeradev/collectd:5.5.0-3

Make sure to add the following lines to your /etc/collectd/collectd.conf:

TypesDB     "/usr/share/collectd/types.db"
TypesDB     "/usr/share/collectd/docker-collectd-plugin/dockerplugin.db"
# ....
<LoadPlugin python>
  Globals true
</LoadPlugin>
# ....
<Plugin python>
   ModulePath "/usr/share/collectd/docker-collectd-plugin"
   LogTraces true
   Interactive false
   Import "dockerplugin"
   <Module dockerplugin>
       BaseURL "unix://var/run/docker.sock"
       Timeout 3
   </Module>
</Plugin>

Now run any other container on the same host: you should be able to see Docker metrics coming to your InfluxDB instance.

[anryko]

Hi @oli-g

Great idea! I'll definitely dig in to it during the weekend.
Thanks!

[anryko]

Didn't succeed in setting up docker plugin to pass statistics to influxdb so far. Encountered reported issue: lebauce/docker-collectd-plugin#37. Will investigate more.

@oli-g does this collectd docker plugin work for you?

[oli-g]

I'm going to try adding the plugin to my collectd Docker image, then I'll get back to you. Thanks for looking into this!

[oli-g]

Here you can find a Collectd Docker image already including the Docker plugin: https://github.com/signalfx/docker-collectd.

They explained their process here: https://signalfx.com/blog/monitoring-docker-at-scale-with-signalfx/.

[anryko]

@oli-g hey,thanks for detailed howto on your setup. It was very helpful. Please checkout initial implementation and let me know what you think.

I encountered the same issue as with processes extended metrics, when for the same metric in some graphs I want to see instances combined and in others separated. Last time I approached it by splitting metric in to two plugins. However, this time the same approach would look ugly. So I added possibility to setup 'multi': (true|fasle) per graph. Seems to work well for this case.

I'm going to rework processes plugin in the same manner. What do you think about that?

[anryko]

Merged docker plugin.

[oli-g]

@anryko Hi! Sorry for not giving you that feedback. I already tested the new Docker plugin integration, and it looks good to me. Just one question: where should I set 'multi': (true|fasle)?

[anryko]

@oli-g Hi! No problem. I added it panel config: https://github.com/anryko/grafana-influx-dashboard/blob/master/getdash.conf.js#L2929

[almereyda]

@oli-g I couldn't find the source files for chimeradev/collectd:5.5.0-3 anywhere on GitHub or the web. Would you mind sharing them for trustability?
Right now it feels wrong to run an unknown container with priviledged access to the docker host.
See ecobytes/collectd-influxdb-grafana-docker@22b8cb6
Additionally I wonder what https://hub.docker.com/r/olinicola/grafana-collectd-dashboards/ intends to provide.