Skip to main content

SaaS and B2B provider

Give each enterprise customer their own complete cluster stack. Dedicated nodes, an isolated cluster, and full cluster-admin access if you choose to offer it. Your platform team handles provisioning and lifecycle. Customers see only their own environment.

Customer access model: External enterprise customers, optionally with cluster-admin on their own cluster. Worker-node model: Private nodes per customer. See Decide whether tenants share hardware.

Typical stack: Platform + private nodes per customer. vMetal optional for physical server fleets. vNode for customers requiring privileged workload support.

What makes this path different: Each customer is fully isolated at the node level. Customers may have cluster-admin rights to their own cluster while remaining invisible to each other and to the platform layer. This path is closely related to the AI Cloud path. The main distinction is whether your customers consume Kubernetes directly (this path) or through your own product API (AI Cloud).

Day 0: Design decisions​

DecisionRead nextOutcome
Define per-customer node allocationPrivate Nodes, Auto NodesEach customer cluster maps to a dedicated node pool. Auto Nodes provisions and reclaims nodes as customers are added or removed.
Define the default network policy postureNetwork policySet a default-deny east-west posture per customer template and confirm the control plane cluster's CNI enforces it. Some CNIs accept NetworkPolicy objects without enforcing them.
Choose the GPU scheduling modelDeviceClasses, DRA ResourceClaimsDevice-plugin-style DeviceClass sync covers simple, whole-GPU allocation. DRA ResourceClaim sync (Pro) supports fine-grained and shared GPU allocation, relevant if customers get cluster-admin access to schedule their own workloads.
Give each customer a tenantTenants, create and configureA tenant is the outer boundary per customer. It determines which control plane clusters, node types, OS images, templates, and connectors they can use. It matters most on this path because customers may access Platform directly rather than only through your product.
Plan customer storageStorage classes from host, CSI storage capacitiesDecide which storage classes each customer can request, and whether to sync host CSIStorageCapacity for scheduler-aware placement.
Plan provisioning automationTemplates, TerraformCustomer cluster creation is triggered by your CRM or provisioning system, not by manual Platform operations.
Define the customer application baselineStacks, create a Stack templateIf every customer receives the same multi-application baseline, define its Apps or Argo CD Applications, dependencies, and health gates once.
Define offboarding and node reclaimManage private nodes, vMetal instance lifecycleDocument how nodes are reclaimed and reimaged when a customer offboards.
Plan security hardening and audit loggingSecurity baseline, CIS hardening guide, Platform audit loggingApply a CIS benchmark baseline to the control plane cluster and decide audit log retention and SIEM integration, especially where customers hold cluster-admin access to their own cluster.

Day 1: Stand up the first customer cluster​

  1. Install vCluster Platform.
  2. Set up a vMetal node provider and node types for your servers or VMs. Auto Nodes uses them to provision and reclaim each customer's nodes.
  3. Create a per-customer cluster template with private nodes, backing store, and HA.
  4. Apply a default-deny network policy in the template and confirm the control plane cluster's CNI enforces it.
  5. Configure per-customer storage classes in the template.
  6. Configure GPU scheduling: sync DeviceClasses for device-plugin allocation, or enable DRA ResourceClaim sync for fine-grained allocation.
  7. Apply the CIS hardening baseline to the control plane cluster and enable Platform audit logging.
  8. If customers receive a standard application baseline, create a StackTemplate and include it in the cluster configuration.
  9. Wire provisioning automation to Platform using Terraform, CLI, or API.
  10. Validate the full customer lifecycle: provision, validate isolation, access as cluster-admin, offboard, and verify node reclaim.

Day 2: Operate​

OperationRead next
Add and remove customersAuto Nodes, provisioning automation
Back up and restore per-customerSnapshots, restore
Upgrade customer clustersUpgrade vCluster
Track GPU utilization and faults for metering and troubleshootingGPU observability templates, NVSentinel GPU observability
Review audit logs and security posturePlatform audit logging, security baseline