Skip to main content

Kubernetes Workload in GKE

This page shows how to deploy a Kubernetes workload on Google Kubernetes Engine (GKE) using the Fractal SDK.

For prerequisites, setup, and deployment file structure, see the main guide.

Java SDK​

public static GoogleKubernetesEngine getGke(String id) {
  return GoogleKubernetesEngine.builder()
    .withId(id)
    .withRegion(EU_WEST1)
    .withNodePools(getNodePools())
    .withK8sWorkload(getK8sWorkload())
    .build();
}

public static CaaSKubernetesWorkload getK8sWorkload() {
  return CaaSKubernetesWorkload.builder()
    .withId("fractal-samples")
    .withDescription("Fractal Service on K8S")
    .withNamespace("fractal")
    .withSSHRepositoryURI("git@github.com:YanchWare/fractal-samples.git")
    .withRepoId("YanchWare/fractal-samples")
    .withBranchName("env/prod")
    .withEnvironmentSecretShortName("my-secret-name") // Add secret access
    // Optional: Use a specific CI/CD profile
    //.withCiCdProfileShortName("my-other-cicd-profile")
    .build();
}

public static Collection<? extends GcpNodePool> getNodePools() {
  return List.of(
    GcpNodePool.builder()
      .withName("nodes")
      .withMachineType(E2_STANDARD2)
      .build()
  );
}

For more details, see the Custom Workload in GKE sample on GitHub.

TypeScript SDK​

The cluster and the workload are two components. The blueprint declares them abstractly — ContainerPlatform and Workload — and the Live System selects Gke and K8sWorkload.

1. Declare the cluster and workload in the blueprint​

import {
  createFractal,
  ContainerPlatform,
  Workload,
} from '@fractal_cloud/sdk/model';

export const fractal = createFractal({
  id: 'gke-workload',
  version: {major: 1, minor: 0, patch: 0},
  description: 'Governed cluster with a custom workload.',
  boundedContextId: {
    ownerType: 'Personal',
    ownerId: process.env['OWNER_ID']!,
    name: 'reusable-templates',
  },
  blueprint: bp => {
    const cluster = bp.add(
      ContainerPlatform({id: 'app-cluster', displayName: 'Application Cluster'})
        .withKubernetesVersion('1.28.5')          // guardrail
        .withNetworkPolicyProvider('calico')      // guardrail
        .withNodePools([
          {name: 'nodes', minNodeCount: 1, maxNodeCount: 3, autoscalingEnabled: true},
        ]),
    );
    const app = bp.add(
      Workload({id: 'app-workload', displayName: 'App Workload'})
        .dependsOn(cluster),
    );
    return {cluster, app};
  },
  operations: s => ({
    /** The container image the app ships. */
    withImage: (image: string) => s.app.set('image', image),
    /** How many replicas to run. */
    withReplicas: (replicas: number) => s.app.set('replicas', replicas),
  }),
});

withNodePools takes {name, minNodeCount?, maxNodeCount?, autoscalingEnabled?} — vendor-neutral capacity, not machine types. Machine sizing is a GCP concept and does not appear on the component.

2. Select the GKE and Kubernetes-workload offers​

import {Gke, K8sWorkload} from '@fractal_cloud/sdk/model';

const liveSystem = fractal
  .specialize()
  .withImage('europe-west1-docker.pkg.dev/my-project/apps/my-app:1.4.0')
  .withReplicas(2)
  .toLiveSystem({
    name: 'gke-workload-dev',
    environment,
    select: {
      'app-cluster': Gke({region: 'europe-west1'}),
      'app-workload': K8sWorkload({namespace: 'default'}),
    },
  });

K8sWorkload is vendor-neutral: it runs on any Kubernetes cluster, so the same selection works against AKS or an on-prem OpenShift cluster. See the Component Reference for every parameter each offer accepts.

For more details, see the TypeScript SDK samples on GitHub.