How Netflix Simplified Batch Compute with Kueue
Netflix transitioned its compute infrastructure to a Kubernetes-native platform using Kueue, enhancing batch job management, capacity utilization, and tenant resource allocation, while maintaining transparency for users and ensuring efficient workload scheduling.
MAIN POINTS
- Netflix migrated from Compute Managed Batch (CMB) to Kueue for improved Kubernetes-native batch job management.
- Kueue supports multi-tenant quota management and integrates with existing Titus scheduling profiles.
- Migration required no changes for end users and maintained container launch rates and throughput.
- Kueue's preemption-based fair sharing allows better resource utilization and prioritization of workloads.
TAKEAWAYS
- Kueue's integration with Titus enhances Netflix's compute platform by modernizing batch abstraction.
- The migration process was simplified by maintaining API parity and addressing complex use cases early.
- Kueue's features, like preemption and fair sharing, improve resource utilization and workload prioritization.
- Netflix's experience with Kueue informs other teams' Kubernetes-native infrastructure developments.