Implementing Out-of-Order Native Histograms in Prometheus
summary
Adds support for out-of-order native histograms in the Prometheus TSDB by teaching the storage to handle histogram encoding types, split chunks on resets, and merge overlapping chunks safely. Changes include separate head vs out-of-order head chunk handling, counter-reset headers on histogram chunks and partial detection of resets at ingest time to avoid expensive query scans. Tests are extended to cover histograms, the PR is under review, and the feature is off by default behind a flag.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:26Prometheus key concepts
- 3:12Out-of-order ingestion basics
- 7:23Native histograms overview
- 10:54Counter resets explained
- 14:48Chunk handling changes
- 15:49Contribution tips
- 18:43Q&A
related talks
25mNative Histograms in Prometheus
36mHow to Use Prometheus's Native Histograms
21mPrometheus Native Histograms in Production
21mSparse High-resolution Histograms in the Prometheus TSDB
04mLightning Talk: Prometheus Native Histograms Are Now a Stable Feature
35mPrometheus Sparse High-Resolution Histograms in Action
26mUsing OpenTelemetry’s Exponential Histograms in Prometheus
22mA New Kid in Histogram Town