0010. Rename tickTime to fxserver_tick_time_seconds

• Status: accepted
• Date: 2026-10-21

Context

/perf/ serves one histogram, tickTime, with a label name that holds the server thread (svMain, svNetwork, svSync); its unit is seconds. In a Prometheus that also monitors other software, tickTime says neither where it comes from nor its unit, and name is the label name that scripts/check_pack.py rejects because elsewhere it would hold a player’s name. A server version may also add metrics to /perf/ that nobody has reviewed.

Considered options

1. Keep only the histogram and rename it in metric_relabel_configs: fxserver_tick_time_seconds_bucket, _sum and _count, with the label thread.
2. Keep the server’s names (tickTime_* with name).
3. Record renamed copies with recording rules and keep the originals.

Decision

Chosen option 1, because the names then follow the Prometheus naming conventions (a prefix for the source, the unit in the name), thread says what the label holds, and the rename happens once, at the scrape, so rules and dashboards use only the new names. The job first keeps only tickTime_bucket, tickTime_sum and tickTime_count, then renames them, copies name to thread and drops name. scripts/check_pack.py fails when a job that scrapes /perf/ lacks the keep rule, the copy or the drop, or copies after the drop, and scripts/smoke_exporters.py checks in CI that exactly the renamed series arrive, with no name label and nothing else from /perf/. Option 3 would store every series twice, and option 2 leaves the generic names in shared dashboards.

Consequences

• Good: rules and dashboards read clearly, and other metrics on /perf/ never enter the database unreviewed.
• Bad: queries written for the server’s own names (for example from other projects) must be translated: tickTime_bucket{name="svMain"} is fxserver_tick_time_seconds_bucket{thread="svMain"}.