A metric that buckets observations by value, the right primitive for tracking latency distributions and computing percentiles.
A histogram is a metric type that records observations into pre-defined value buckets, e.g. a request-latency histogram with buckets at 10ms, 50ms, 100ms, 500ms, 1s, 5s. Each bucket is a counter of how many observations fell into it. Histograms are the right primitive for distributions because they let the backend compute any percentile (p50, p95, p99, p99.9) cheaply at query time. Prometheus, OpenMetrics, and OpenTelemetry all support histograms natively.
Average and gauge metrics hide the tail, the slow 1% of requests that hurt users most. Histograms make the tail visible and queryable. Switching latency dashboards from average to p99 (computed from a histogram) is one of the highest-impact observability changes a team can make, and the histogram is what enables it.
See the part of the platform that handles histogram in production.