Deploy the kubeflow-kfserving helm chart on AWS EKS

Question

Pulumi · Accepted Answer

To deploy the Kubeflow KFServing Helm chart on AWS EKS, we will need to accomplish a few high-level steps:

1. Set up an AWS EKS cluster using the `eks.Cluster` resource from Pulumi's EKS package.
2. Deploy the Kubeflow KFServing Helm chart within the EKS cluster using Pulumi's Kubernetes provider and the `kubernetes.helm.v3.Chart` resource.

The below TypeScript program performs the following actions:

- Configure an EKS cluster, specifying the desired settings for node groups, such as instance type and min/max size.
- Define a Helm chart resource for Kubeflow KFServing, specifying the necessary Helm values according to the chart's requirements.

Make sure you have AWS credentials configured and the Kubernetes/Helm Pulumi providers installed.

Here's the TypeScript program that sets up the EKS cluster and deploys KFServing:

```typescript
import * as eks from "@pulumi/eks";
import * as k8s from "@pulumi/kubernetes";
import * as pulumi from "@pulumi/pulumi";

const projectName = "kubeflow-kfserving";

// Create an EKS cluster with the desired configuration
const cluster = new eks.Cluster(projectName, {
    instanceType: "t2.medium", // Use a medium sized instance
    desiredCapacity: 2,        // Desired number of instances in the node group
    minSize: 1,                // Minimum number of instances in the node group
    maxSize: 3,                // Maximum number of instances in the node group
    // other cluster configurations such as VPC, IAM roles, etc.
});

// Export the cluster's kubeconfig so that we can access it
export const kubeconfig = cluster.kubeconfig;

// Initialize a Kubernetes Provider using the EKS cluster's kubeconfig
const k8sProvider = new k8s.Provider(projectName, {
    kubeconfig: cluster.kubeconfig.apply(JSON.stringify),
});

// Define the Kubeflow KFServing Helm chart
const kfservingChart = new k8s.helm.v3.Chart("kfserving", {
    chart: "kfserving",
    version: "0.6.0", // Specify the version of KFServing you wish to deploy
    fetchOpts: {
      repo: "https://kubeflow.github.io/kfserving/", // The Helm chart repository for Kubeflow KFServing
    },
    // Include any custom values required for the Helm chart
    values: {
        // Custom values for KFServing or other dependencies
    },
}, { provider: k8sProvider });

// Export the Helm chart name
export const kfservingChartName = kfservingChart.metadata.apply(m => m.name);

// When the program is deployed, you should get kubeconfig as output and
// will be able to use it to connect to your new EKS cluster and verify the deployment.
```

This program is a starting point and you may need to fine-tune the configuration. For example, the instance type, and the minimum and the maximum size of instances might need adjustment based on the workload. You should also provide any required custom Helm values in the 'values' field to satisfy KFServing's configuration needs. Check the [KFServing Helm chart](https://github.com/kubeflow/kfserving/tree/master/charts/kfserving) for further customization options.

To apply this Pulumi program:

1. Save the code in a file with a `.ts` extension (e.g., `index.ts`).
2. Run `pulumi up` in the same directory as the file.
3. Pulumi CLI will perform the deployment, and you will be provided with outputs such as the `kubeconfig`.