Company: Uber Technologies
Industry: Travel & Transportation
Uber is a global technology company that revolutionized the transportation industry with its ride-hailing platform. Today, it operates in over 70 countries, providing services that include food delivery (Uber Eats), freight logistics, and autonomous mobility. With millions of rides and deliveries happening daily, Uber’s platform relies on a vast microservices architecture that demands high performance, scalability, and efficiency
Challenge
Uber operates at a massive scale, with over 5,000 microservices, 11,000 weekly commits, and 500,000+ deployments per week. This complexity leads to inefficiencies across Uber’s tech stack, causing performance bottlenecks, high infrastructure costs, and incident troubleshooting challenges. Historically, Uber’s approach to optimization was reactive, requiring engineers to investigate issues manually after they had already impacted performance. The company needed a proactive, automated, and scalable approach to detect and resolve inefficiencies before they caused problems.
Solution
Uber transitioned from reactive profiling (manual incident-based profiling) to automated, proactive profiling with Grafana Cloud Profiles. Initially, engineers used U Monitor, an ad-hoc manual profiler, which evolved into Auto Profiler, triggering profiling at high CPU usage, and then Periodic Profiling, collecting daily snapshots. With Pyroscope’s continuous profiling, Uber enabled always-on, low-overhead monitoring, storing data in Grafana Cloud for historical analysis. Engineers now use differential flame charts to detect regressions and tagging/filtering for deeper insights. To further improve performance, Uber integrated AI-powered code optimization, which identifies inefficiencies, generates fixes, and validates improvements with machine-learning benchmarks. This automation has already delivered 100+ AI-assisted optimizations, making Uber’s infrastructure more cost-effective and high-performing.
“With AI-powered profiling with Grafana Cloud Profiles and the resulting automated fixes, we’re democratizing performance optimization, making expert-level improvements accessible to all developers.”
Milind Chabbi, Senior Staff Researcher, Uber
Impact:
- Cost savings: Pyroscope helped Uber identify a 30% allocation overhead in a backend service, leading to a 10,000-core reduction—saving millions of dollars in infrastructure costs.
- Faster incident resolution: Engineers used Pyroscope to diagnose a memory leak in 30 minutes, a process that previously took days.
- Developer productivity: AI-powered profiling automates inefficiency detection and provides suggested fixes allowing engineers to spend less time debugging and more time building.
- Scalability without overhead: Uber’s initial six-month POC across six Tier 1 services showed no visible overhead allowing them to roll out continuous profiling across all services by the end of 2025.
“Pyroscope helped us pinpoint and resolve a memory leak in just 30 minutes—something that used to take days of manual debugging.”
Shauvik Roy Chaudhary, Engineering Manager, Uber
Your guides
