Better Histograms for Prometheus

Explore the evolution and potential improvements of histograms in Prometheus in this 30-minute conference talk by Björn Rabenstein from Grafana Labs. Delve into the current limitations of Prometheus histograms, particularly their high per-bucket cost, and understand why adopting established techniques from academic publications and other metrics-processing systems has been challenging. Discover the speaker's latest findings on breaking through barriers to implement efficient high-resolution histograms in Prometheus. Learn about conventional histograms, cumulative histograms, sparse histograms, and their impact on storage space. Gain insights into the accumulating entropy problem and potential solutions for better resolution and aggregation in Prometheus histograms.

Intro
Isolation
Recap
Prometheus Histograms
Wishlist
Conventional Histograms
Histograms in Prometheus
The accumulating entropy
Cumulative histograms
Sparse histogram
Storage
Storage Space
Wish List
True Spoiler
Resolution
Aggregation
Outro