Skip to content

Setup

This feature enables one-click Kubernetes cluster deployment using Rafay-managed bare metal VMs, designed specifically for GPU PaaS customers. The solution removes the complexity of provisioning virtual machines and installing Kubernetes by offering a self-service, single-click deployment process.

Two key user personas are supported:

  • PaaS Admin: Creates and customizes Kubernetes compute profiles, defining the hardware (CPU/GPU, memory, storage) and configuration options available for deployment
  • End User: Selects a pre-defined compute profile via the Developer Portal, provides minimal details such as cluster name and worker node count, and launches the cluster

The approach is intentionally simplified, focusing on user experience rather than backend configurations such as environment templates.

The system also supports flexible setups, including the option to run CPU-based control plane nodes with GPU-based worker nodes, depending on available inventory.

User Roles

PaaS Admin

  • Manages Kubernetes compute profiles
  • Customizes compute profile configurations based on the target end user
  • Hides complex configurations from end users, exposing only essential fields like the number of worker nodes

Navigation: PaaS Studio → Compute Profiles → New Compute Profile) → Edit Input Settings

Hierarchy

  • The PaaS Admin defines and customizes configuration parameters in the Input Settings screen.
    • Can enable the “Override” option in Input Settings, allowing end users to modify specific parameters during cluster creation
  • Profiles are then saved and made available for end users to select during cluster creation

Hierarchy

End User

  • Navigate to Developer Hub -> Compute -> Kubernetes -> New Kubernetes
  • Selects the appropriate shared Kubernetes compute profile

Hierarchy

  • Specifies essential fields such as Name, Description, and editable parameters (editable only if the PaaS Admin has enabled Override for those fields)
  • Click Deploy to deploys the cluster with a single click

Hierarchy

  • User receives a kubeconfig file to interact with the deployed cluster

Hierarchy