Prometheus Cardinality Troubleshooter

This is a triage guide for when Prometheus is actively on fire: OOMing, queries timing out, or your Grafana Cloud bill just doubled. It walks you through the tsdb status endpoints, shows you how to drill into per-metric and per-label breakdowns, and points you at the usual suspects (unbounded path labels, trace_ids leaking into metrics, histogram explosions). The best part is the opening rule: it explicitly warns you not to labeldrop distinguishing labels at scrape time, which looks like it fixes cardinality but actually merges distinct series and silently breaks rate() calculations. If you've ever stared at a cardinality fire and wondered where to start clicking, this gives you the runbook.

Install to Claude Code

npx -y skills add grafana/skills --skill prometheus-cardinality-troubleshooter --agent claude-code