Skip to content

Requirements

The following requirements must be met to provision and operate clusters using the One-Click GPU PaaS Kubernetes deployment model.

Configuration Parameters (PaaS Admin)

PaaS Admins define the following in compute profiles:

  • Kubernetes Version: Pre-selected in the profile
  • Control Plane Node Configuration:
    • CPU-only or mixed CPU/GPU setup
    • Node count (e.g., 1, 3, or 5 for HA)
  • Worker Node Configuration:
    • GPU-enabled or CPU-only nodes
    • Node SKU (CPU cores, memory, GPU type)
    • Storage configuration
  • CNI Plugin: Defaults to Calico; can be changed in the profile
  • Networking & Security: Proxy settings, public IP allocation, firewall rules
  • Blueprint Selection: Standard or GPU-enabled blueprint for workloads and monitoring
  • Advanced Settings Exposure: Controls which fields are visible to end users

Input Parameters (End User)

End users provide only minimal inputs when launching a cluster:

  • Cluster Name: Unique name for the new Kubernetes cluster
  • Compute Profile: Select from pre-published shared profiles created by the PaaS Admin
  • Worker Node Count: Number of worker nodes to provision
  • (Optional) Cluster Labels: Add custom labels for workload grouping
  • Deployment Action: Single-click Deploy initiates provisioning

All other settings (CPU/GPU type, storage, network, node sizing) are pre-configured in the compute profile by the PaaS Admin.

Infrastructure Requirements

  • Compute Profiles: At least one published profile must be available to end users
  • Rafay Inventory Integration: Bare metal or VM resources must exist and match profile specifications
  • GPU Inventory: For GPU workloads, GPU-equipped nodes must be available

Operational Constraints

  • Master node count and compute profile selection are immutable after cluster creation
  • All nodes within a cluster must match the SKU defined in the selected compute profile
  • Public IP assignments and dedicated control plane settings are fixed post-deployment