Automating efficiency at Uber scale with Grafana Cloud Profiles and Assistant

Uber’s microservice environment comprises myriad services with different reliability tiers, developed in different languages, and spanning stateful and stateless categories. In such a system spanning several million CPU cores, performance visibility and code-level attribution are of prime importance to ensure efficient resource utilization, prevent regressions, and improve customer experience. 

In this session, Senior Researcher Chris Zhang from Uber's Programming Systems Group shares how the team deployed always-on profiling with Grafana Cloud for their complex production systems in order to unlock these efficiency gains while remaining cost-effective and scalable. Along the way, they deployed Grafana Assistant in various pipelines, driving measurable impact across the business.

Now that profiles are accessible to thousands of engineers –  giving them visibility to expensive code – Chris describes the next major leap. The team now continuously and automatically optimizes the profiles with little human oversight, while ensuring very high confidence using Generative AI coupled with neurosymbolic analyses advancing modern trends in agentic systems. The combination of continuous profiling and continuous optimization has democratized code optimization at Uber, delivering leaps in cost-effectiveness and customer experience.