One Controller to Rule Them All - Taming Multiple Orchestrator... Srikar Paruchuru & Egor Grishechko

This talk covers the challenges of managing multiple orchestrators at scale within Uber's Compute platform, which supports over one million workloads across 200 Kubernetes clusters in two regions. The speakers, Srikar Paruchuru and Egor Grishechko, introduce the Service Scale abstraction, a custom resource definition (CRD) and controller designed to resolve conflicts between different orchestrators dictating target scales for workloads. They share insights on their controller's design, its advantages over alternatives like the scale sub-resource, and the significant cost savings achieved through optimized regional failovers. Additionally, they discuss the challenges faced during the migration to the new controller, including issues such as controller races and event overload, along with valuable lessons learned from the process.