Reference · Glossary
Histogram
Last updated
A histogram metric buckets observed values (like latency) so you can estimate percentiles such as p50 and p95.
#When to use
Production AI and API services where you need visibility, safer deploys, or verified inbound callbacks.
#When not to
Local one-off notebooks with no users. Do not collect high-cardinality labels “just in case.”
#Practical tips
- Prefer a few well-named metrics over dozens of vanity charts
- Alert on symptoms users feel (errors, latency) before chasing every gauge
- Keep runbooks next to dashboards
#Learn next
- Lesson: `ai-monitoring`
- Related: prometheus, latency, observability