Ceph RadosGW Usage Exporter

Prometheus exporter that scrapes Ceph RadosGW usage information (operation, buckets, and user). This info is gathered from RGW using the Admin Operations API.

Requirements

Ceph RGW must have the admin API enabled via config. (It is enabled by default)

rgw enable apis = "s3, admin"

This exporter requires a user that has the following capabilities:
- buckets=read
- users=read
- usage=read (only if you enable the usage log. See below)
If using a loadbalancer in front of RGW, please make sure your timeouts are set appropriately. Clusters with a large number of buckets or large number of users+buckets could cause the usage query to exceed the loadbalancer timeout

Optional

To get operations usage metrics, you must enable the usage log

rgw enable usage log = true

Configuration

All inputs / config is done via ENV variables

Variable	Default	Required?	Description
RGW_EXPORTER_PORT		Required	The URL of the RadosGW instance to scrape (example: https://objects.example.com/)
RGW_EXPORTER_RGW_URL		Required	The URL of the RadosGW instance to scrape (example: https://objects.example.com/)
RGW_EXPORTER_ACCESS_KEY		Required	S3-style access key of the user to use for scraping
RGW_EXPORTER_SECRET_KEY		Required	S3-style secret key of the user to use for scraping
RGW_EXPORTER_LOG_LEVEL	"info"		The log level to use [debug, info, warn, error, fatal]
RGW_EXPORTER_INTERVAL	"1m"		How often to scrape ceph. NOTE: This is a minimum duration between scrapes. If a scrape takes longer than the interval, multiple scrapes will not overlap. The current scrape will finish and then immediately start a new scrape

Usage

This exporter is published as a docker container: ghcr.io/richiesams/radosgw-exporter:v<tag>

You can run the container directly

docker run --rm \
    -e RGW_EXPORTER_RGW_URL=https://objects.example.com/ \
    -e RGW_EXPORTER_ACCESS_KEY=MY_ACCESS_KEY \
    -e RGW_EXPORTER_ACCESS_KEY=MY_SECRET_KEY \
    -p 8080:8080 \
    $(IMAGENAME):$(TAG)

Or via your favorite orchestration system

Metrics will be exposed at the /metrics endpoint

Final notes