Pyroscope 2.0: Continuous profiling that scales

With Pyroscope 2.0, we’re delivering a ground-up rearchitecture of our open source continuous profiling database. Designed to make continuous profiling faster and more cost-effective at scale, the latest major release of Pyroscope includes a new storage layer: replication-free, no persistent disks, and reads fully decoupled from writes.

This talk covers why we rebuilt, the architectural decisions behind 2.0—including how symbol co-location exploits profiling data's natural redundancy for a 20x storage reduction—and what we measured after rolling it out across every Grafana Cloud region.

In a demo, we'll investigate a page for request timeouts on Pyroscope's own query-backend service, using a span heatmap to spot the slow requests, drilling into an individual profile, and pinpointing exactly which code path is causing the timeouts.