Automating efficiency at Uber scale with Grafana Cloud Profiles and Assistant
Date:
Conference talk at ObservabilityCON 2026 · San Francisco
Over the past two years, I’ve worked with colleagues at Uber and Grafana on always-on profiling infrastructure across Uber’s 6,000+ services. In this talk, I shared how we built it and what we learned along the way.
The talk covers why random sampling can miss smaller services, how a central control plane manages collection without blocking applications, and how we use profiles to investigate CPU use, memory allocation, and lock contention. It also includes lessons from failures and the effects of performance fixes on other parts of a system.
The final part covers Grafana Assistant, profile access through MCP, and a design for an automated optimization loop that combines program analysis, generative AI, and checks for equivalent behavior.
Recording coming soon.