Scaling Kubernetes to Thousands of Nodes Across Multiple Clusters, Calmly
Talkat KubeCon + CloudNativeCon North America 2019 · on Tuesday, 2019-11-19· at 16:25· Ballroom Sec 20CD - San Diego Convention Center, Upper Level·
summary
Airbnb moves from a few handcrafted clusters at ~600 nodes to tens of clusters totaling >7,000 nodes by adopting a multi‑cluster model, hitting practical limits (etcd instability, scheduler latency) and fixing them with etcd v3 and higher quota backend bytes, simplified anti‑affinity and Kubernetes upgrades. Cluster creation becomes automated with Chef and Terraform generators plus Helm umbrella charts and kube‑gin, and regeneration prevents drift so teams reduce cluster creation time from over a week to under an hour while capping clusters at ~400 nodes to avoid large‑cluster failure modes.
AI-written summary · about
chapters (12)
- 0:00Start
- 3:02Monolith
- 4:41One Touch
- 5:28Migration
- 10:16Multicluster considerations
- 11:24We Got Lucky
- 17:21What is a cluster?
- 18:36One Deployable Unit
- 20:16Deplay Process
- 22:04Clusters and Cluster Types
- 22:35Cluster Launch
- 25:15Some things are easy to change
related talks
03mFrom 5 to 1,300+ Clusters: Declarative Scaling for Private Kubernetes
15mKubernetes and etcd Features That Unlocked 15k Node Clusters
32mDon't Start With 500 Clusters: Patterns in Scaling Multi-Cluster Kubernetes…
26mThe Road To Cilium: Migrating 150+ Kubernetes Clusters at Airbnb
28mScaling in Kubernetes Safely on On-Prem KaaS Across 1,300+ Clusters and 40,000+ Nodes
56mManaging from 1 to 1 million Kubernetes Clusters
42m2000 Nodes and Beyond: How We Scaled Kubernetes to 60,000-Container Clusters
34mEmbracing Upstream Kubernetes in Web Scale Organization