From 5 to 1,300+ Clusters: Declarative Scaling for Private Kubernetes
summary
Kubernetes becomes the infrastructure control plane by modeling clusters as custom resource definitions with controllers that create and manage OpenStack VMs, automating provisioning, scaling, upgrades, and recovery so cluster creation drops from weeks to hours and operations become zero‑downtime. LY scales from 5 to 1,300+ clusters, 40,000+ nodes and ~1M containers while 15 platform engineers support the fleet (roughly one engineer per ~90 clusters), and automatic replacement handles node failures without human intervention.
AI-written summary · about
related talks
31mDeclarative Underlays: Scaling Purpose-Built Infrastructure(Clusters)...
30mScaling Kubernetes to Thousands of Nodes Across Multiple Clusters, Calmly
42m2000 Nodes and Beyond: How We Scaled Kubernetes to 60,000-Container Clusters
28mA Brief History of Kubernetes Fleet Controllers & Essential Features
34mPanel: Improving and Managing Kubernetes at Scale
56mManaging from 1 to 1 million Kubernetes Clusters
35mTo Infinite Scale and Beyond: Operating Kubernetes Past the Steady State
37mKubernetesception: How to run a 1000 control planes