How Netflix Simplified Batch Compute with Kueue

What this post added

This post details the migration of Netflix's custom batch compute solution (CMB) to Kueue, an open-source Kubernetes-native job queueing system. It explains the motivations for the migration, including the limitations of CMB and the advantages of Kueue (e.g., native preemption, integration with existing schedulers). The post outlines the migration strategy, emphasizing transparency for end-users and zero regressions in performance. It describes the architectural changes, mapping CMB tenants to Kueue's Cohorts, ClusterQueues, and LocalQueues. Key lessons learned, such as maintaining API parity and migrating complex use cases early, are shared. The current state highlights the successful rollout of Kueue managing millions of jobs, with a focus on enhanced fair sharing and preemption features leading to increased resource utilization.