Kcp: Towards 1,000,000 Clusters, Name^WWorkspaced CRDs
summary
kcp turns the apiserver into a partitioned control plane that offers workspaces — cheap, millisecond-created logical clusters implemented by passing a logical cluster name and using it as an etcd key prefix — and exposes CRD-like schemas via APIExport/APIBinding so one service can serve up to 1,000,000 workspaces. It shards apiservers for horizontal scale, uses a globally replicated distributed cache for eventual consistency, provides a controller model with local plus global informers, and runs real workloads on underlying Kubernetes clusters via sync agents.
AI-written summary · about
chapters (13)
- 0:00kcp vision
- 4:59kcp compute model
- 7:30kcp extensions
- 16:01Generic workspaces API sharing
- 23:06Getting started kcp workspaces
- 27:26Reconstructing compute in kcp
- 31:46kcp community resources
- 31:59Q&A hierarchical namespaces
- 33:34Global scale consistency in kcp
- 35:41Workspace implementation
- 36:47Multi-cluster compute service
- 38:13Multi-workspace controllers
- 39:21kcp scalability trade-offs
related talks
05mkcp: Scaling The Kubernetes Control Plane For The Multi-Clust...
32mBuilding a Platform Engineering API Layer with kcp
38mScaling Kube-Apiserver to Thousands of Tenants – Lessons Learned From kcp
34mBuilding a Platform Engineering API Layer with KCP
27mBuilding a Platform Engineering API Layer with kcp
01h 15mTutorial: Exploring Multi-Tenant Kubernetes APIs and Controllers With Kcp
22mTaming the VM Sprawl: KCP Meets KubeVirt
43mPlatform Engineering Redefined: Evolving Kubernetes Beyond Container Orchestration