feat(e2e): Run e2e tests in kube

.github/actions/setup-kind-cluster/action.yaml

@@ -0,0 +1,187 @@
+name: Setup Kind cluster with NeMo Platform
+description: >
+  Creates a Kind cluster, installs tooling (kind, kubectl, Helm, uv),
+  pre-pulls images, deploys NeMo Platform via Helm, and waits for the
+  API to become healthy.
+
+inputs:
+  kind_cluster_name:
+    description: Kind cluster name
+    required: true
+  kube_namespace:
+    description: Kubernetes namespace for NeMo Platform
+    required: false
+    default: nemo-platform
+  kube_gateway_name:
+    description: Gateway resource name
+    required: false
+    default: nmp-e2e-gateway
+  image_registry:
+    description: Container image registry
+    required: true
+  image_tag:
+    description: Container image tag
+    required: true
+  helm_values:
+    description: Helm values file path (relative to repo root)
+    required: false
+    default: e2e/k8s/values/kind.yaml
+  kind_image_pull_token:
+    description: Token for pulling images into Kind
+    required: true
+  kind_image_pull_user:
+    description: Username for pulling images into Kind
+    required: true
+  ngc_api_key:
+    description: NGC API key (can be a placeholder for CPU-only)
+    required: false
+    default: not-used
+
+outputs:
+  cluster_url:
+    description: The NeMo Platform cluster URL
+    value: ${{ steps.export-url.outputs.cluster_url }}
+
+runs:
+  using: composite
+  steps:
+    - name: Free disk space
+      uses: ./.github/actions/free-disk-space
+      with:
+        disable_swap: "true"
+        remove_haskell: "true"
+        remove_java: "true"
+        remove_ruby: "true"
+        remove_swift: "true"
+        prune_docker: "true"
+
+    - name: Install kind
+      shell: bash
+      env:
+        KIND_VERSION: v0.32.0
+      run: |
+        set -euo pipefail
+
+        case "$(uname -m)" in
+          x86_64) arch=amd64 ;;
+          aarch64|arm64) arch=arm64 ;;
+          *)
+            echo "Unsupported architecture: $(uname -m)" >&2
+            exit 1
+            ;;
+        esac
+
+        kind_url="https://github.com/kubernetes-sigs/kind/releases/download/${KIND_VERSION}/kind-linux-${arch}"
+        curl -fsSLo "${RUNNER_TEMP}/kind" "${kind_url}"
+        curl -fsSLo "${RUNNER_TEMP}/kind.sha256sum" "${kind_url}.sha256sum"
+        sed "s#  kind-linux-${arch}#  ${RUNNER_TEMP}/kind#" "${RUNNER_TEMP}/kind.sha256sum" | sha256sum -c -
+        sudo install -m 0755 "${RUNNER_TEMP}/kind" /usr/local/bin/kind
+
+    - name: Install kubectl
+      shell: bash
+      env:
+        KUBECTL_VERSION: v1.33.7
+      run: |
+        set -euo pipefail
+
+        case "$(uname -m)" in
+          x86_64) arch=amd64 ;;
+          aarch64|arm64) arch=arm64 ;;
+          *)
+            echo "Unsupported architecture: $(uname -m)" >&2
+            exit 1
+            ;;
+        esac
+
+        kubectl_url="https://dl.k8s.io/release/${KUBECTL_VERSION}/bin/linux/${arch}/kubectl"
+        curl -fsSLo "${RUNNER_TEMP}/kubectl" "${kubectl_url}"
+        curl -fsSLo "${RUNNER_TEMP}/kubectl.sha256" "${kubectl_url}.sha256"
+        echo "$(cat "${RUNNER_TEMP}/kubectl.sha256")  ${RUNNER_TEMP}/kubectl" | sha256sum -c -
+        sudo install -m 0755 "${RUNNER_TEMP}/kubectl" /usr/local/bin/kubectl
+
+    - name: Install Helm
+      uses: azure/setup-helm@dda3372f752e03dde6b3237bc9431cdc2f7a02a2 # v5.0.0
+
+    - name: Install uv
+      uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7.6.0
+      with:
+        python-version: "3.13"
+        enable-cache: true
+        version-file: pyproject.toml
+        cache-dependency-glob: uv.lock
+
+    - name: Start kind cluster
+      shell: bash
+      env:
+        KIND_CLUSTER_NAME: ${{ inputs.kind_cluster_name }}
+        KUBE_NAMESPACE: ${{ inputs.kube_namespace }}
+        NGC_API_KEY: ${{ inputs.ngc_api_key }}
+      run: bash e2e/k8s/scripts/setup_local_kind_cpu.sh
+
+    - name: Set default kubectl namespace
+      shell: bash
+      env:
+        NAMESPACE: ${{ inputs.kube_namespace }}
+      run: kubectl config set-context --current --namespace="${NAMESPACE}"
+
+    - name: Verify Gateway API setup
+      shell: bash
+      env:
+        NAMESPACE: ${{ inputs.kube_namespace }}
+        KUBE_GATEWAY_NAME: ${{ inputs.kube_gateway_name }}
+      run: |
+        set -euo pipefail
+        kubectl wait --for=condition=Established crd/gateways.gateway.networking.k8s.io --timeout=2m
+        kubectl wait --for=condition=Established crd/httproutes.gateway.networking.k8s.io --timeout=2m
+        kubectl get gatewayclass cloud-provider-kind
+        kubectl -n "${NAMESPACE}" get gateway "${KUBE_GATEWAY_NAME}"
+
+    - name: Pre-pull GHCR images into kind
+      shell: bash
+      env:
+        KIND_IMAGE_PULL_TOKEN: ${{ inputs.kind_image_pull_token }}
+        KIND_IMAGE_PULL_USER: ${{ inputs.kind_image_pull_user }}
+        NMP_E2E_REGISTRY: ${{ inputs.image_registry }}
+        NMP_E2E_TAG: ${{ inputs.image_tag }}
+      run: |
+        e2e/k8s/scripts/prepull_kind_images.sh \
+          "${NMP_E2E_REGISTRY}/nmp-api:${NMP_E2E_TAG}" \
+          "${NMP_E2E_REGISTRY}/nmp-core:${NMP_E2E_TAG}" \
+          "${NMP_E2E_REGISTRY}/nmp-cpu-tasks:${NMP_E2E_TAG}"
+
+    - name: Install NeMo Platform
+      shell: bash
+      env:
+        NAMESPACE: ${{ inputs.kube_namespace }}
+        NMP_E2E_REGISTRY: ${{ inputs.image_registry }}
+        NMP_E2E_TAG: ${{ inputs.image_tag }}
+        HELM_VALUES: ${{ inputs.helm_values }}
+        REQUIRE_NMP_E2E_IMAGES: "true"
+        POSTGRES_IMAGE: docker.io/library/postgres
+        BUSYBOX_IMAGE: docker.io/library/busybox
+      run: |
+        if ! e2e/k8s/scripts/install_helm_e2e.sh; then
+          echo "--- helm list -A ---"
+          helm list -A || true
+          echo "--- helm status ${NAMESPACE}/nemo-platform ---"
+          helm status -n "${NAMESPACE}" nemo-platform || true
+          echo "--- kubectl get all -n ${NAMESPACE} ---"
+          kubectl get all -n "${NAMESPACE}" || true
+          exit 1
+        fi
+
+    - name: Export cluster URL
+      id: export-url
+      shell: bash
+      env:
+        NMP_E2E_CLUSTER_URL: ${{ env.NMP_E2E_CLUSTER_URL }}
+      run: |
+        echo "cluster_url=${NMP_E2E_CLUSTER_URL}" >> "${GITHUB_OUTPUT}"
+
+    - name: Wait for API
+      shell: bash
+      env:
+        NMP_E2E_CLUSTER_URL: ${{ env.NMP_E2E_CLUSTER_URL }}
+      run: |
+        test -n "${NMP_E2E_CLUSTER_URL}"
+        e2e/k8s/scripts/wait_for_api.sh "${NMP_E2E_CLUSTER_URL}/cluster-info" 120