Lightning Talk: Learning from Cortex to Improve Promscale HA
summary
Promscale replaces fragile DB locks with a Cortex-inspired design that deduplicates Prometheus HA replicas by storing replica identity in the samples and coordinating writes with immutable leases per time range so only one replica owns that interval. Data flows: two (or more) Prometheus replicas add external labels, send through a load balancer to a Promscale tier, which uses the lease to decide which samples to ingest into TimescaleDB on Postgres. This removes one-to-one coupling, improves scalability, and gives a durable history of who owned each interval.
AI-written summary · about
chapters (7)
- 0:00Intro
- 0:13What is Prometheus High-Availability
- 1:24Original Solution
- 2:04Problem 1
- 3:04New Approach: immutable leases
- 4:01Using an insight from Cortex
- 5:13New Architecture
related talks
31mWhen Prometheus Can’t Take the Load Anymore
26mWhen Prometheus Can’t Take the Load Anymore
11mLightning Talk: Beyond Billions: Operating Thanos, Prometheus & OpenTelemetry...
32m“Gluing it all together” aka building a platform around Prometheus
13mLightning Talk: Observability-First DevSecOps: Building Resilient Multi-Cloud P...
10mUpstream-First, High Scale Prometheus Ecosystem
24mBanking Observability at Scale
04mLightning Talk: Prometheus Rules management and validation