Expand description
Gauges summarising a sliding window of observations.
QuantileGauge exposes a small fixed set of latency quantiles as a
<name>{quantile="..."} gauge series, computed at scrape time from a
~2-minute sliding window, instead of a full bucketed histogram. It lets a
dashboard read <name>{quantile="0.5"} directly in place of
histogram_quantile(0.5, rate(<name>_bucket[2m])), collapsing the
per-bucket series (and, for QuantileGaugeVec, the per-label bucket
expansion) down to one series per quantile.
The summaries are computed on each node over its own observations, so they cannot be re-aggregated across nodes in PromQL; query them per host.
Structsยง
- Quantile
Gauge - A single latency distribution exposed as
<name>{quantile="..."}. - Quantile
Gauge Vec - A latency distribution kept per value of one label, exposed as
<name>{<label>="...", quantile="..."}. One window is created lazily per observed label value.