collectd is a daemon which collects system performance statistics periodically and provides mechanisms to store the values in a variety of ways, for example in RRD files.

This image allows you to run collectd in a completely containerized environment, but while retaining the ability to report statistics about the host the collectd container is running on.

The SignalFx collectd Docker plugin is also included, and will report metrics about your containers.

Run collectd with the default configuration with the following command:

docker run --privileged \
  --net="host" \
  -e "SF_API_TOKEN=XXXXXXXXXXXXXXXXXXXXXX" \
  -v /etc/hostname:/mnt/hostname:ro \
  -v /proc:/mnt/proc:ro \
  -v /:/hostfs:ro \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /etc:/mnt/etc:ro \
  quay.io/signalfuse/collectd

If you don't want to report Docker metrics from the SignalFx Docker collectd plugin, then you can leave out the argument that mounts the Docker socket -v /var/run/docker.sock:/var/run/docker.sock.

docker run --privileged \
  --net="host" \
  -e "SF_API_TOKEN=XXXXXXXXXXXXXXXXXXXXXX" \
  -v /etc/hostname:/mnt/hostname:ro \
  -v /proc:/mnt/proc:ro \
  -v /:/hostfs:ro \
  -v /etc:/mnt/etc:ro \
  quay.io/signalfuse/collectd

If you just want to look around inside the image, you can run bash. Then you can run /opt/setup/run.sh yourself to configure the bind mount and start collectd.

docker run -ti --privileged \
  --net="host" \
  -e "SF_API_TOKEN=XXXXXXXXXXXXXXXXXXXXXX" \
  -v /etc/hostname:/mnt/hostname:ro \
  -v /proc:/mnt/proc:ro \
  -v /:/hostfs:ro \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /etc:/mnt/etc:ro \
  quay.io/signalfuse/collectd bash

If you don't want to pass your API token through a command-line argument, you can put SF_API_TOKEN=XXXXXXXXXXXXXXXXXXXXXX into a file (that you can chmod 600) and use the --env-file command-line argument:

docker run --privileged --env-file=token.env \
  --net="host" \
  -v /etc/hostname:/mnt/hostname:ro \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /proc:/mnt/proc:ro \
  -v /:/hostfs:ro \
  -v /etc:/mnt/etc:ro \
  quay.io/signalfuse/collectd

If you want to apply custom dimensions and values to all metrics, you set the environment variable DIMENSIONS to a string of space separated Key/Value pairs defined KEY=VALUE.

docker run --privileged --env-file=token.env \
  --net="host" \
  -e DIMENSIONS="hello=world foo=bar"
  -v /etc/hostname:/mnt/hostname:ro \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /proc:/mnt/proc:ro \
  -v /:/hostfs:ro \
  -v /etc:/mnt/etc:ro \
  quay.io/signalfuse/collectd

On CoreOS because /etc/*-release are symlinks you want to mount /usr/share/coreos in place of /etc.

docker run --privileged \
  --net="host" \
  -e "SF_API_TOKEN=XXXXXXXXXXXXXXXXXXXXXX" \
  -v /etc/hostname:/mnt/hostname:ro \
  -v /proc:/mnt/proc:ro \
  -v /:/hostfs:ro \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /usr/share/coreos:/mnt/etc:ro \
  quay.io/signalfuse/collectd

Yes. Collectd needs access to the parent host's /proc filesystem to get statistics. It's possible to run collectd without passing the parent host's /proc filesystem without running the container as privileged, but the metrics would not be accurate.

Yes. Setting the --net option to "host" allows the container to see the host's networking stack and provide accurate information.

It's actually optional but we use this to get the host OS's version so the SignalFx Collectd Plugin will function correctly and report the right information. If you leave it out you will see the container's OS in the meta-data for this host.

This is a requirement for the df plugin to correctly report information about the host and not from the container. If you explicitly disable the df plugin then this is no longer a requirement to run the container.

You have an option to either bind mount your host's /etc/hostname to the container's /mnt/hostname or to set the environment variable COLLECTD_HOSTNAME. You may set the environment variable using the runtime option -e "COLLECTD_HOSTNAME=<hostname>" where <hostname> is a valid host name string. You may also set the environment variable in an optional environment file. Please note that if /etc/hostname is mounted and COLLECTD_HOSTNAME is set, that the mounted host name will be used. A hostname must be specified regardless of the method.

Because the docker container's proc filesystem has been replaced by the host and because the PIDs are remapped, they no longer line up. See the example above on how to run the image with a different command so you can then execute /.docker/run.sh yourself to be on it while it runs.

Yes! You are required to set the SF_API_TOKEN and provide a hostname either by mounting /etc/hostname to /mnt/hostname or setting COLLECTD_HOSTNAME, but you also can set the following:

Click this link to read about Extending SignalFx Docker collectd image