Docker Stats exporter for Prometheus

This exporter uses Apocas's Dockerode library.

This exporter was created because the official Docker exporter only exports engine metrics not containers', and the alternative Google's cAdvisor uses quite a lot of resources to run.

The output essentially mirrors the CLI output of docker stats.

Usage

Arguments

--port     9487         Exporter listens on this port (default = 9487)
--interval 15           Polling interval in seconds (default = 15, minimum 3)
--hostip   127.0.0.1    Docker engine IP to connect to (if using HTTP)
--hostport 2375         Docker engine port to connect to (if using HTTP)
--collectdefault        Collect default Prometheus metrics as well (default = false)

If no hostip and hostport provided, it defaults to connect via socket to /var/run/docker.sock.

Environment Variables

The arguments can also be set as env variables instead. Useful if you're using it in a Docker container.

1. DOCKERSTATS_PORT
2. DOCKERSTATS_INTERVAL
3. DOCKERSTATS_HOSTIP
4. DOCKERSTATS_HOSTPORT
5. DOCKERSTATS_DEFAULTMETRICS

Installation

From Source

Node.js 14 is required to run it.

git clone git@github.com:wywywywy/docker_stats_exporter.git
cd docker_stats_exporter
npm ci
npm start

With Docker

docker run -d --restart=always -p 9487:9487 -v /var/run/docker.sock:/var/run/docker.sock wywywywy/docker_stats_exporter:latest

Prometheus Config

Add this to prometheus.yml and change the IP/port if needed.

- job_name: 'docker_stats_exporter'
    metrics_path: /
    static_configs:
    - targets:
        - '127.0.0.1:9487'

Notes on Metrics

If you don't see any memory metrics, try to run docker stats to see if the memory usage ever goes above zero. If not, that means cgroup memory support is not enabled in the kernel, which is most likely the reason if you're using Raspberry Pi OS.

See this issue for instructions to have it enabled: moby/moby#18420

Notes on Performance

Because of the way docker stats works, it always takes at least 2 seconds to output the results. Basically it takes a snapshot, then after a second it takes another snapshot to compare the results.

So there is no point pooling the Docker engine more than once every 3 or so seconds.

Also because docker stats runs one process for each container to gather metrics, this exporter is not suitable for hosts that have a large number (hundreds) of containers.

TODO

1. Block IO metrics
2. Other useful metrics not in docker stats

Contributing

Yes, contributions are always welcome.
Fork it, clone it, submit a pull request, etc.

License

This is licensed under the Apache License 2.0.