hard20 pts · first answer countsprometheusmonitoring

Broken p95 latency query

This PromQL query is meant to show p95 latency per service, but it returns no useful data. What is wrong?

histogram_quantile(0.95,
  sum(rate(http_request_duration_seconds_bucket[5m])) by (service)
)

A.rate() must not be used on bucket metricsB.The quantile must be written as 95, not 0.95C.histogram_quantile only works on summary metricsD.The sum drops the le label; it must be sum by (le, service)

← Securing a role for a SaaS vendor Fixing alert fatigue with SLOs →