Deep Dive: Cortex
summary
Cortex is a Prometheus-compatible, multi-tenant TSDB that stores samples in compressed “chunks” (roughly six-hour blobs a few KB each) and writes them to object/cloud stores (S3/GCS/BigTable/DynamoDB). It replicates data 3-way for no-gap HA, tolerates node loss in the demo, and serves queries via an inverted index plus chunk read path using batched iterators (batch ~12) to avoid OOMs. Performance relies on an index cache, large chunk memcache (hundreds of GB), and a query front-end that splits by day and yields ~50 ms cache hits, ~500 ms 99th pctl.
AI-written summary · about
chapters (14)
- 0:00Intro
- 4:30Demo
- 12:58Storage Engine
- 14:53Query Path
- 17:28Inverted Index
- 23:38Index Cache
- 25:33Query Frontend
- 27:54Performance Results
- 28:41Closing Comments
- 29:17Open Source
- 29:49Peter Savio
- 33:25Bryan Boreham
- 35:32Global distributed parameters
- 36:52Multitenant
related talks
28mCortex 101: Horizontally Scalable Long Term Storage for Prometheus
20mCortex: Intro and Production Tips
25mCortex: Ingesting and Querying 100s of Billions of Samples Per Day
22mCortex: Ingesting and Querying 100s of Billions of Samples Per Day
45mScaling Prometheus with Cortex, one time series at a time
16mCortex: Multi-tenant Scalable Prometheus
41mWeave Cortex: Horizontally Scalable, Multitenant Prometheus as a Service
20mIntro to Scaling Prometheus with Cortex