Prometheus Docker Exporter

Exports stats from Docker API (/containers/<id>/stat endpoint).

Metrics are collected on each request, everything in single thread. Fetching stats from Docker API for lots of containers on a small machine could take few seconds, but this implementation will not use host CPU excessively (looking at you, cAdvisor!).

Supported metrics

• cpu
• block i/o (cgroups v2)
• memory
• network

Usage

usage: glorpen-docker-exporter [-h] [-a ADDR] [-p PORT] [--sysfs SYSFS]

options:
  -h, --help            show this help message and exit
  -a ADDR, --addr ADDR  address to lsiten on, defaults to 0.0.0.0
  -p PORT, --port PORT  port to listen on, defaults to 8080
  --sysfs SYSFS, -s SYSFS
                        path to sysfs to use, defaults to /sys

Access to /sys is needed for resolving block device names (/sys/dev/block).

Exporter needs r/w access to Docker socket (eg. /run/docker.sock).

Supported Docker envs:

• DOCKER_HOST: The URL to the Docker host.
• DOCKER_TLS_VERIFY: Verify the host against a CA certificate.
• DOCKER_CERT_PATH: A path to a directory containing TLS certificates to use when connecting to the Docker host.

Docker

docker run -v /sys:/sys:ro -v /run/docker.sock:/run/docker.sock:rw glorpen/prometheus-docker-exporter:1.0.0

Pypi

pip install glorpen-docker-exporter
python -m glorpen-docker-exporter

Example Prometheus queries

CPU usage per container (as in docker stat command):

delta(container_cpu_seconds_total[2m]) / delta(container_cpu_system_seconds_total[2m]) * avg_over_time(container_cpu_online[2m])

Container HDD reads and writes:

rate(container_blkio_reads_bytes_total[2m]) > 0
rate(container_blkio_writes_bytes_total[2m]) > 0