Skip to content

Setup

One-Click Serverless Pods

This feature enables one-click deployment of serverless pods on Kubernetes clusters using Rafay-managed GPU and CPU resources.

It eliminates the need for manual pod provisioning and configuration by providing a self-service, single-click experience for developers to launch Jupyter notebooks, inference services, or custom workloads.

Follow these steps to create and launch a new Serverless Developer Pod.

  • In the controller, navigate to Compute Type -> Developer Pods and click New Developer Pod. The Pod Catalog opens

ℹī¸ Note:
Developer Pod templates are designed for reusability. By adjusting configuration parameters such as resources, pod images, or SKUs, the same template can be applied to multiple use cases (e.g., Jupyter, VSCode Server, inference workloads) without requiring additional development effort.

  • Choose Serverless Developer Pods from the available options and click Select to continue

Select Pod Type

  • Enter the following details in the configuration form:

Configure Pod

Provide a name for the pod and select the workspace where the pod should be deployed.

Pod Details

Resources

Specify the CPU, memory, number of GPUs, and GPU model to use with the pod.

Pod Resources

Select the desired GPU model from the dropdown list.

GPU Model

Select Pod Image

Pods can come pre-installed with frameworks and libraries. This allows data scientists to start using the pod immediately, without spending time installing software or managing version conflicts.

Select Pod Image

Select a Pod Image from the list of available options in the dropdown.

Pod Images

Cost Estimate

A real-time cost estimate is provided based on the selected pod configurations.

Pod Cost Estimate

Info

Pods are metered by the hour, with a minimum usage of 1 hour.

View Pods

Multiple pods can be active at any given time. By selecting the Pods menu on the left, users can view a list of all pods across their workspaces.

View Pod

Viewing Pod Runtime Details

  • Navigate to the Clusters page and select the required cluster. Under the Resources tab, select the namespace to view the list of pods. QoS Class displays the Quality of Service for the pod. In this example, the pod is marked as Guaranteed since the resource limits and requests are identical.
  • Click on the Describe option for the pod to view runtime details. The Describe view provides detailed information such as:
    • Resource Limits and Requests: CPU and memory limits are identical to the requests, ensuring guaranteed compute availability
    • Mounts: Displays the mounted volumes. In this example, the SSH keys are mounted inside the pod through the volume ssh-key-volume
    • Node Selector: Shows that the pod is pinned to a specific node with GPU access. This ensures that the pod is scheduled only on that node
  • To view the service details, select Services for the required namespace. Here, a service of type NodePort is available. This service exposes a port, enabling connectivity to the application.

This view provides a comprehensive runtime overview of pod configuration, scheduling constraints, and operational state directly from the Controller UI.

Access Pod

To access a pod, click on its name. A view like the example below will be displayed. Pods can also be accessed from a laptop using SSH.

Access Pod

Accessing Developer Pods via SSH

Developer Pods can be accessed through SSH using automatically generated credentials and commands.

Step 1: Generate Private Key File

When a pod is created, the system provides a command to generate a private key file.

This command not only creates the file but also applies the correct permissions, so no additional configuration is needed.

echo "-----BEGIN RSA PRIVATE KEY-----
<key-contents>
-----END RSA PRIVATE KEY-----" > <pod-name>-ssh.key
chmod 600 <pod-name>-ssh.key

Step 2: Connect to the Pod via SSH

Use the provided SSH command to connect directly to the Developer Pod.

ssh -i <pod-name>-ssh.key ubuntu@<pod-ip> -p <port>

Sample Output

Welcome to Ubuntu 24.04.2 LTS (GNU/Linux 6.8.0-1028-oracle x86_64)

Step 3: Verify the Environment

Once inside the pod, standard Linux commands can be executed to validate access.

Example:

uname -a

Sample Output

Linux <pod-name>-deployment-xxxxxx 6.8.0-1028-oracle #29-Ubuntu SMP Fri Jun 20 19:33:55 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux

This confirms that the Developer Pod is accessible and ready for use.

Delete Pod

To delete a pod, click the ellipsis on the far right of the selected pod and choose Delete.

Delete

Info

Once deletion is initiated, it cannot be stopped or reversed. A new pod can be created if needed.