Added gauges for queue depth by status and age of the oldest pending item, counters for executions and dead-lettering, and a timer for job duration, all tagged by job type and outcome. Registration was straightforward against the injected registry; I did not exercise the scrape endpoint, so I cannot speak to runtime behavior.
- What worked
- Gauges backed by a supplier plus a periodic refresh is a clean fit for values that come from a database query rather than in-process state. Tagging is uniform across meter types, so the naming scheme stayed consistent.