KubeInception: running up to a thousand clusters in the real Cloud-Native way
summary
Kamaji is an operator that runs control planes as pods inside a management cluster, uses Cluster API for worker VMs, and treats etcd as a sharded multi-tenant datastore (kine/Postgres or etcd shards) so dozens to thousands of control planes share backing databases. Benchmarks show ~100 control planes in 15 minutes and 1,000 in about an hour, with network/CPU spikes accounted for; costs drop ~70% versus one-VM-per-control-plane EKS, and control planes support scale-to-zero and constant reconciliation.
AI-written summary · about
chapters (9)
- 0:00KubeInception intro
- 3:15Thousands of clusters challenges
- 7:50Kamaji for multi-tenancy
- 12:03Kamaji architecture
- 18:03Demo: Kamaji control plane
- 24:31Scaling to 1k clusters
- 32:21Benchmarks and cost savings
- 34:52When to use Kamaji
- 35:55Q&A
related talks
34mScaling Kubernetes up to 1,000 Control Planes - Dario Tranchitella
40mMay the Control Plane be with you: Kamaji and the rise of Kubernetes at scale
37mKubernetesception: How to run a 1000 control planes
26mRunning Kubernetes Clusters the Kubernetes-Native Way
35mKubernetes-in-Kubernetes with Kamaji, KubeVirt, and the Cluster API
56mManaging from 1 to 1 million Kubernetes Clusters
37mManaging and Running Multiple Kubernetes Clusters in Hybrid Setups
34mPanel: Improving and Managing Kubernetes at Scale