Prometheus Native Histograms in Production
Talkat Observability Day Europe 2023 · on Tuesday, 2023-04-18· at 09:50· Hall 7, Room E | Ground Floor | Europe Complex·
summary
Native Prometheus histograms run behind a feature flag and keep one series per histogram instead of many bucket series, cutting series from ~16,000 to 964 in the test. With a 100 buckets per histogram cap the deployment sees about ~21,000 populated buckets total, giving roughly 10× resolution for half the storage compared with classic histograms. Resets happen infrequently (worst case eight resets in two weeks), common quantile queries are as fast or faster, and heatmaps gain markedly higher fidelity.
AI-written summary · about
chapters (4)
- 0:00Start
- 10:13Frequency of resets and resolution reduction
- 16:22Prometheus resource consumption
- 17:39Query performance
related talks
25mNative Histograms in Prometheus
36mHow to Use Prometheus's Native Histograms
35mPrometheus Sparse High-Resolution Histograms in Action
05mNative Histograms With Custom Buckets
04mLightning Talk: Prometheus Native Histograms Are Now a Stable Feature
23mImplementing Out-of-Order Native Histograms in Prometheus
42mPrometheus Intro and Deep Dive
22mA New Kid in Histogram Town