Choose your starting point
A complete platform stacks four layers. Machines run a Kubernetes cluster. That cluster runs vCluster Platform and becomes the control plane cluster. Platform then delivers clusters and other services to your tenants.
You don't have to build every layer. Start at the first layer you don't already have, and reuse what you run today.
Building for your own engineering teams rather than for customers? The same layers apply, and your tenants are your internal teams. See Team and CI clusters as a service for that path.
| What you already run | Start here | What you can skip |
|---|---|---|
| Bare metal and nothing else | Bare metal only | Nothing |
| A machine provisioner or inventory | A provisioner or machine inventory | Building your own provisioning |
| One or more Kubernetes clusters | Kubernetes clusters | The machine layer and the first Kubernetes cluster |
| Clusters created with vCluster | Clusters created with vCluster | Re-creating those clusters |
Bare metal only
Use one bare metal server to bootstrap the control plane cluster, then bring the rest of your bare metal in through vMetal.
- Install vCluster Standalone on one bare metal Linux server. Standalone is a complete Kubernetes distribution that runs as a single binary, with no existing cluster required. See the Standalone quick start.
- Install vCluster Platform on that cluster with Helm, as a separate step. See Install with Helm.
- Connect the rest of your bare metal servers through a vMetal provisioning driver, such as Metal3 for bare metal you provision directly. See Get started with vMetal.
A provisioner or machine inventory
If you already manage bare metal or VMs with another system, vMetal can provision through it instead of replacing it.
- Install vCluster Platform on any Kubernetes cluster. That cluster becomes the control plane cluster. See the Platform quick start.
- Create a node provider for the system you run. vMetal supports these provisioning drivers:
- Metal3, for bare metal servers you provision directly
- NVIDIA Infra Controller (NICo), for servers already managed at a NICo site
- KubeVirt, for VMs on a KubeVirt cluster you operate
- Terraform, for cloud VMs and anything else a Terraform provider can create
- Choose what the machines become. A machine can be a cluster node, host a cluster control plane, or stay an instance with no cluster that a tenant reaches over SSH.
Kubernetes clusters
If you already run Kubernetes, install Platform on one of your clusters and connect the others.
- Install vCluster Platform on one cluster, with the CLI or with Helm. That cluster becomes the first control plane cluster. See the Platform quick start.
- Connect your other clusters as additional control plane clusters with
vcluster platform add cluster. See Connect a cluster. - Give tenant workloads private nodes. Join machines to each cluster as private nodes yourself, or let Auto Nodes provision them on demand. See Capacity models.
Clusters created with vCluster
If you already run clusters with the vCluster CLI or Helm chart, bring them under Platform management instead of re-creating them.
- Install vCluster Platform if you haven't yet. See the Platform quick start.
- Import each cluster with
vcluster platform add vcluster. Add--allto import the clusters in every namespace of the control plane cluster. For a Standalone cluster, usevcluster platform add standaloneinstead. See Add existing clusters. - Decide how much Platform manages. By default, an imported cluster stays externally managed instead of being managed by the platform agent. To have the agent manage it, connect its control plane cluster first, then import with
--external=false. See Platform-managed and externally deployed clusters.
Next steps
- Decide whether tenants share hardware before you offer clusters to anyone outside your trust boundary.
- Capacity models explains how tenant workloads get compute.
- Roles and scopes explains who builds and administers each layer.
- The service pages, such as Kubernetes as a service, show the full build order for each offering.