Explore Airbnb's journey to heterogeneous Kubernetes clusters in this 35-minute conference talk from KubeCon + CloudNativeCon. Discover the technical and organizational challenges faced when migrating from homogeneous to heterogeneous clusters, aimed at improving cost-efficiency and workload optimization. Learn about the infrastructure changes required, from cluster provisioning to API modifications, and the strategies employed to address customer concerns and build confidence in the new operating model. Gain insights into performance testing, pod topology spread, cluster autoscaler modifications, and scheduling simulations. Understand the benefits of this transition, including future-proofing and cost savings, and explore next steps and existing pull requests in the Kubernetes ecosystem.
What Kind of CPU is it Anyways? Airbnb's Journey to Heterogeneous Clusters
CNCF [Cloud Native Computing Foundation] via YouTube
Overview
Syllabus
Introduction
Outline
Brief History of Kubernetes
Nobody knows whats going on
The unhappy path
Single Instance Type
Single Instance Type Modification
Pause Pod Modification
Custom Expander Plugin
Organizational Issues
Migration Fatigue
Customer Concerns
Performance Testing
Very Latency Testing
Technical Issues
Pod Topology Spread
Logs
Empty ASGS
Cluster Autoscaler
How does Scheduling Simulation work
Topology Zone Key
Short Term Fix
Summary
Biggest Benefit
Future Proof
Cost
Next Steps
Existing PRs
Conclusion
Thank You
Taught by
CNCF [Cloud Native Computing Foundation]
