How Prometheus indexes Data and Why You Should Care
Talkat PrometheusDay Europe 2022 · on Tuesday, 2022-05-17· at 12:00· Pavilion 3, Room F | Level 2 | Central Forum + Live Stream, Feria Valencia, Spain + Live Stream·
summary
Explains that Prometheus stores data in 2 hour blocks and uses a postings index mapping label -> series and a series index mapping series -> chunks so query execution first picks overlapping blocks then intersects postings then scans per-series chunks and seeks samples, so cost multiplies with number of blocks and series. Every new label value creates a new series so high cardinality explodes work and disk seeks per chunk are the main expense. Mentions Promscale using TimescaleDB as an alternative backend.
AI-written summary · about
chapters (10)
- 0:00Start
- 1:33Prometheus Data Model
- 1:56WHAT ARE SERIES?
- 2:21WHAT PROMETHEUS INDEX CARES ABOUT
- 3:01ANATOMY OF A BLOCK
- 4:53Series Index
- 6:14WHAT IS A QUERY?
- 10:26For a more complex query
- 13:04What is High-cardinality?
- 13:42Under high label cardinality
related talks
30mIntro: Prometheus
46mPrometheus as a Time Series Database
16mPrometheus 101
37mPrometheus Deep Dive
33mPrometheus Project Intro
01h 19mIntro + Deep Dive: Prometheus
39mHow To Take Prometheus Planet Scale: Massively Large Scale Metrics Installations
37mJourney inside Prometheus: Architecture, updates, outlook