0017. Build age and the Recommended build

• Status: accepted
• Date: 2026-10-21

Context

The alerts asked for are “build older than 8 weeks” and “behind the Recommended build”. The game server does not say when its build was made or released: info.json has the build number in the version string, which a label cannot compare with a number. The fxserver role of the luminare.gameserver collection writes two textfile metrics for node_exporter: the build number that runs, and the time its current link was switched. The platform publishes the current Recommended build in a public version list, which json_exporter can read. Its support policy (a Recommended build is supported until 6 weeks after the next one, builds older than 3 months cannot be joined from the server list) makes both alerts worth having.

Considered options

1. Age from the switch time (fxserver_artifact_installed_timestamp_seconds), and the Recommended build from the version list through an optional json_exporter job, compared with fxserver_artifact_build.
2. The build date of each build, looked up in the version list.
3. A list of build numbers and dates kept in this repository.

Decision

Chosen option 1, because both inputs exist today and are numbers PromQL can compare. The switch time is a lower bound of the build’s age: a build that was already old when it was switched on fires later. “Behind” means older than every Recommended build of the last 14 days (min_over_time of the version list), so a new Recommended build gets two weeks before the alert fires, and an outage of the version list does not end the alert. Option 2 is not possible: the list has support dates but no release dates. Option 3 would go stale between releases of this pack.

Consequences

• Good: both alerts are tested over 17 simulated days (tests/rules/build.test.yml), including an update, a newer build than the Recommended one and an outage of the version list. The version list module was checked against the real list on 2026-10-21.
• Bad: both alerts need the textfile metrics, so they work out of the box only with the collection (or a script that writes the same file). The version list job adds an outbound request from the Prometheus host, every 4 minutes; no alert watches that job, so a broken list stays unnoticed until someone looks at up{job="fxserver_versions"}. FXServerArtifactOld also fires when the build that runs is still the current Recommended build; the runbook says to silence it then.