This Prometheus exporter exposes the status of the Puppet Agent on the host it is running on.

Unlike puppet-prometheus_reporter and other solutions that rely on the Puppet Agent successfully a report to the Puppet Server, this allows you to actively monitor the status of every node in your environment that is discoverable by Prometheus.

# HELP puppet_config Puppet configuration.
# TYPE puppet_config gauge
puppet_config{environment="",server="puppet.redacted"} 1
# HELP puppet_last_catalog_version The version of the last attempted Puppet catalog.
# TYPE puppet_last_catalog_version gauge
puppet_last_catalog_version{version="1680640107"} 1
# HELP puppet_last_run_at_seconds Time of the last Puppet run.
# TYPE puppet_last_run_at_seconds gauge
puppet_last_run_at_seconds 1.6806401024160552e+09
# HELP puppet_last_run_duration_seconds Duration of the last Puppet run.
# TYPE puppet_last_run_duration_seconds gauge
puppet_last_run_duration_seconds 28.023470087
# HELP puppet_last_run_success 1 if the last Puppet run was successful.
# TYPE puppet_last_run_success gauge
puppet_last_run_success 1

groups:
  - name: Puppet
    rules:
      - alert: MultiplePuppetFailing
        expr: count(puppet_last_run_success) - count(puppet_last_run_success == 1) > 4
      - alert: PuppetEnvironmentSet
        expr: puppet_config{environment!=""}
        for: 4h
      - alert: PuppetFailing
        expr: puppet_last_run_success == 0
        for: 40m
      - alert: LastPuppetTooLongAgo
        expr: time() - puppet_last_run_at_seconds > 3*60*60
        for: 40m

We've found it worthwhile to avoid alerting on conditions affecting individual nodes until that condition has persisted long enough to affect more than one run. (For example, a transient hiccup in an APT proxy, etc...)

Alerting on the catalog being too old helps catch situations like a node being set to an environment that no longer exists, or if it's having TLS or network issues contacting the Puppet Server.

Alerting on a non-default environment being set helps catch operator error, for example when a node is used to test changes from a branch environment but forgotten about after that branch is merged.

This is an open-source release of something we've used for quite a while internally.

• The package APIs functions are likely to be refactored drastically, potentially without warning in release notes.

  (This isn't really meant to be a library.)

• We may add a systemd unit to the packaging at some point.

• We probably won't make breaking changes to the arguments without warning.

• Test coverage
• General code hygiene and refactoring
• Better packaging (supporting RPM distros, including a systemd unit, etc...)

Contributions considered, but be aware that this is mostly just something we needed. It's public because there's no reason anyone else should have to waste an afternoon (or more) building something similar, and we think the approach is good enough that others would benefit from adopting.

This project is licensed under the Apache License, Version 2.0.

Please include a Signed-off-by in all commits, per Developer Certificate of Origin version 1.1.