The percentile distributions exposed through commons monitoring are currently exposed as Prometheus' Summary metric type.
These percentiles are calculated by the instrumented process per unique combination of label values for each Summary type metric. Once calculated in this way, it is not possible to calculate percentiles over a metric for all values of one or more labels.
For example, given a metric which exposes the 99.9th percentile response time for a given request type, it is possible to know this value per server but not across an entire cluster.
Prometheus' Histogram metric type depends on calculations being made by the Prometheus server. This allows Prometheus to calculate percentiles per instance, across an entire cluster, or for any combination of label values sets of interest.
- Commons Monitoring instrumentation library has been updated to support Prometheus' Histogram metric type
- Commons Monitoring instrumentation library exposes means of settings Histogram bucket boundaries
- Commons Monitoring instrumentation library continues to support Summary metric type