Prometheus Marathon task count exporter

This exporter for Prometheus only exposes one metric (fow now), the number of tasks for each app in a Marathon cluster. It has one dimension, the name of the app.

This exporter is not meant to report low-level metrics on Marathon. If you are looking for this, you should look at the Mesos expoer or the Marathon exporter.

Example output

marathon_task_count{task="app-01.production"} 10
marathon_task_count{task="app-02.production"} 3
marathon_task_count{task="app-03.staging"} 0

Note : the task name format was taken from the Mesos exporter for consistency. If you have an app in /cluster/production/app-01, it will be reported as app-01.production.cluster.

Build

$ go get github.com/apognu/marathon_task_exporter
$ go build github.com/apognu/marathon_task_exporter

Usage

$ ./marathon_task_exporter -help
Usage of ./marathon_task_exporter:
  -marathon.url string
        Marathon instance URL
  -web.listen-address string
        Address to listen on for HTTP interface (default ":9091")
  -web.telemetry-path string
        Path under which to expose metrics (default "/metrics")

$ ./marathon_task_exporter -marathon.url=http://my.marathon.tld

Alerting

You can use Prometheus's Alert Manager with this exporter, by first grouping by the task label in your route configuration:

route:
  group_by: [ 'task' ]

An example alert rule could look like this:

ALERT ProductionAPIInstanceCount
  IF marathon_task_count{task = "/production/.+"} < 3
  FOR 30s
  ANNOTATIONS {
    summary = "Support Marathon Task count",
    description = "Number of tasks for *{{$labels.task}}* has changed -> *{{$value}}*"
  }

What's next?

• Try to include Mesos labels as metric dimensions
• Take app health checks into account to count only live tasks
• Handle HTTP authentication on Marathon