0013. Probed target in the instance label

• Status: accepted
• Date: 2026-10-21

Context

The jobs fxserver_json, fxserver_blackbox and fxserver_tunnel scrape json_exporter and blackbox_exporter, which fetch the real target per scrape. Without a relabel rule, instance would be the exporter’s address (127.0.0.1:7979) for every target. A fixed alias per job (for example game-01-json) would keep every probed URL and address out of the labels, which matters because dashboards made from this data get shared. Every series already carries the server label, a short name chosen per game server, and module, which says which probe or endpoint made it.

Considered options

1. Copy the probed target (__param_target) into instance, as the blackbox_exporter and json_exporter examples do.
2. A fixed alias per job.
3. An alias built from server and module.

Decision

Chosen option 1, because when a probe fails, up == 0 and the probe series then name the URL or port that failed, without a look into the scrape config. Option 2 hides that, and option 3 only repeats two labels that every series already has. The shipped targets are loopback addresses and example.com; the address of a real server or admin hostname appears in instance only in the operator’s own Prometheus. Shared views must not show it: rules, alerts and dashboards built on this data group by server and module, never by instance.

Consequences

• Good: a failing target is named in the series itself.
• Bad: instance holds a host address or a hostname. The later dashboard step must keep it out of shared views (no instance in panels, legends, variables or tables), its lint must fail on any query or legend that uses instance, and the share copy must not contain it.