The Road To Cilium: Migrating 150+ Kubernetes Clusters at Airbnb
summary
Airbnb migrates 150+ Kubernetes clusters (13,000+ nodes) from AWS VPC CNI to Cilium over eight months, validating at scale with Kubeburner (100 pods/s) and control-plane tests. Key findings: pod start time stays fast with IP pre-allocation, node bootstrap is ~50% slower due to extra init work, and AWS EC2 API throttling—not Cilium—is the primary bottleneck; prefix delegation reduces EC2 mutating calls by ~60%, rollout uses Argo, node-level CNI selection, and Cilium’s AWS operator while disabling advanced features until stable.
AI-written summary · about
chapters (13)
- 0:00Cilium migration at Airbnb
- 1:07Cilium vs VPC CNI choice
- 3:17Kubernetes fleet scale & timeline
- 3:58Cilium control plane performance
- 6:53Prefix delegation reduces EC2 API calls
- 9:20EC2 API calls on node bootstrap
- 13:16EC2 API scalability improvements
- 14:41Safe cluster rollout strategy
- 16:07Node vs pod CNI migration controls
- 19:03Worker node migration first
- 20:48Node-level rollout choice
- 21:30Production migration issues
- 24:05Q&A
related talks
32mSwitching your CNI — Simone Sciarrati & Federico Hernandez
24mConfluent's Multi-Cloud Journey to Cilium: Pitfalls and Lessons Learned
30mScaling Kubernetes to Thousands of Nodes Across Multiple Clusters, Calmly
25mMore Churn No Problem: Lessons Learned Running Cilium in Production
35mPerforming Infrastructure Migrations at Scale
28mLive Migrating Production Clusters From Calico to Cilium
24mHigh-Scale Networking for ML Workloads With Cilium
16mCilium Wars: Scaling Up Your Network with Jedi-like Cilium Handling Techniques - Yash Tandon