Scaling Prometheus with Cortex, one time series at a time
summary
Cortex claims a horizontally scalable, highly available long‑term backend for Prometheus that actually works: the demo sustains 70k samples/s demo and a cited production cluster handles 2.5M samples/s over 24h, with a default replication factor 3 and in‑memory recent data plus periodic 2‑hour block uploads to an object‑store (S3/GCS/MinIO). Sharding uses a hash ring, queries parallelize across blocks and ingesters, compactor deduplicates replicated blocks, and reads show ~100 ms 99th‑percentile latency while writes report ~5 ms 99th.
AI-written summary · about
chapters (16)
- 0:00Start
- 3:15Introduction
- 5:23Multi-Tenancy
- 7:03Introduction about Cortex
- 7:19What's Cortex
- 8:00Single Binary Mode and the Block Storage
- 8:41Cortex Overview
- 11:32Demo
- 12:18Prometheus High Availability
- 17:07Scaling Up a Cortex Cluster
- 20:37Architecture and Storage
- 22:59Compactor
- 23:56Microservices Architecture
- 24:54Sharding and Replication
- 33:12What's Next in Cortex
- 39:24Supported Object Stores
related talks
20mIntro to Scaling Prometheus with Cortex
37mIntro: Cortex
25mCortex: Ingesting and Querying 100s of Billions of Samples Per Day
39mHow To Take Prometheus Planet Scale: Massively Large Scale Metrics Installations
39mDeep Dive: Cortex
41mWeave Cortex: Horizontally Scalable, Multitenant Prometheus as a Service
20mCortex: Intro and Production Tips
37mPrometheus Deep Dive