Fluentd Docker Metrics Input Plugin

This is a Fluentd plugin to collect Docker metrics periodically.

How it works

It's assumed to run on the host server. It periodically runs docker ps --no-trunc -q to get a list of running Docker container IDs, and it looks at /sys/fs/cgroups/<metric_type>/docker/<container_id>/ for relevant stats. You can say this is an implementation of the metric collection strategy outlined in this blog post.

Installing

to be uploaded on Rubygems

Example config

<source>
  type docker_metrics
  stats_interval 1m
</source>

Parameters

• stats_interval: how often to poll Docker containers for stats. The default is every minute.
• cgroup_path: The path to cgroups pseudofiles. The default is sys/fs/cgroup.
• tag_prefix: The tag prefix. The default value is "docker"

Example output

2014-06-26 18:16:43 +0000 docker.memory.stat: {"key":"memory_stat_total_active_anon","value":26025984,"source":"docker:precise64:b7f17c393775476bc0999cb6dcb4c6416e94b0473317375b9a245985dc6e91c5"}
2014-06-26 18:16:43 +0000 docker.memory.stat: {"key":"memory_stat_total_inactive_file","value":131072,"source":"docker:precise64:b7f17c393775476bc0999cb6dcb4c6416e94b0473317375b9a245985dc6e91c5"}

In particular, each event is a key-value pair of individual metrics. Also, it has a field whose value is "<tag_prefix>:<hostname>:<container_id>"