Pushing the Limits of Prometheus at Etsy - Scaling Beyond Performance Boundaries

This conference talk explores the journey of pushing Prometheus beyond its performance limits at Etsy. Dive into an insider's perspective on scaling a single Prometheus instance using a 128-core machine with 4TB of RAM that processed up to 500 million metrics at peak. Learn about the challenges encountered in Prometheus' design and how they were overcome, techniques for combining observability signals (metrics, profiles, and traces) to identify and resolve performance bottlenecks, and strategies for optimizing metrics volume while enhancing reliability under load. Discover practical lessons and actionable takeaways from operating one of the industry's largest Prometheus servers to build a more resilient observability stack. The presentation is delivered by Chris Leavoy from Etsy and Bryan Boreham from Grafana Labs as part of a CNCF event.