test-e2e-lifecycle.yml

name: E2E Tests Lifecycle

on:
  workflow_dispatch:
  pull_request:
    paths:
      - 'cmd/vmcp/**'
      - 'cmd/thv-operator/**'
      - 'pkg/**'
      - 'test/e2e/thv-operator/**'
      # This job is the only one that runs the Ginkgo suite under
      # test/e2e/thv-operator, and that suite sources every backend container
      # image from test/e2e/images. A change there alters what these tests
      # actually deploy, so it has to trigger them.
      - 'test/e2e/images/**'
      - '.github/workflows/test-e2e-lifecycle.yml'

# Cancel superseded PR runs; keep other events independent.
concurrency:
  group: test-e2e-lifecycle-${{ github.event.pull_request.number || github.run_id }}
  cancel-in-progress: ${{ github.event_name == 'pull_request' }}

permissions:
  contents: read

jobs:
  e2e-test-lifecycle:
    name: E2E Test Lifecycle
    runs-on: ubuntu-8cores-32gb
    timeout-minutes: 30
    env:
      YARDSTICK_IMAGE: ghcr.io/stackloklabs/yardstick/yardstick-server:1.2.0
    defaults:
      run:
        shell: bash
    strategy:
      fail-fast: false
      matrix:
        version: [
          "kindest/node:v1.33.7",
          "kindest/node:v1.34.3",
          "kindest/node:v1.35.1"
        ]

    steps:
    - name: Checkout code
      uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
      with:
        persist-credentials: false

    - name: Disable containerd image store
      # Workaround for https://github.com/kubernetes-sigs/kind/issues/3795
      # Docker 29+ defaults to containerd image store, which causes
      # `kind load docker-image` to fail for multi-arch images because
      # `docker save` preserves the OCI index referencing all platforms
      # even when only the host platform layers were pulled.
      # --platform on docker pull is not sufficient; the image store
      # itself must be switched back to the classic overlay2 driver.
      run: |
        sudo mkdir -p /etc/docker
        echo '{"features":{"containerd-snapshotter": false}}' | sudo tee /etc/docker/daemon.json
        sudo systemctl restart docker

    - name: Set up Helm
      uses: azure/setup-helm@9bc31f4ebc9c6b171d7bfbaa5d006ae7abdb4310 # v5.0.1
      with:
        version: v4.3.1

    - name: Setup Ko
      uses: ko-build/setup-ko@61b4d1d396f5b2e7d6bb6fefdce3dc38d1a13445 # v0.10

    - name: Set up Go
      uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
      with:
        go-version: 'stable'
        cache: true

    - name: Install Task
      uses: arduino/setup-task@c0bc642852239c2689f73f4ea6459c29405f3c52 # v3.0.0
      with:
        version: 3.44.1
        repo-token: ${{ secrets.GITHUB_TOKEN }}

    - name: Create KIND Cluster with port mappings
      uses: helm/kind-action@1544676c07bb3570fcb70166beaa07c4528fd4aa # v1.15.1
      with:
        cluster_name: toolhive
        version: v0.33.0 # kind
        config: test/e2e/thv-operator/kind-config.yaml
        node_image: ${{ matrix.version }}

    - name: Setup cluster and install CRDs
      run: |
        kind get kubeconfig --name toolhive > kconfig.yaml
        export KUBECONFIG=kconfig.yaml
        task operator-install-crds

    - name: Build and load test images
      run: |
        # Build and load vmcp image
        echo "Building vmcp image..."
        VMCP_IMAGE=$(KO_DOCKER_REPO=kind.local ko build --local -B ./cmd/vmcp | tail -n 1)
        echo "Loading vmcp image ${VMCP_IMAGE} into kind..."
        kind load docker-image --name toolhive ${VMCP_IMAGE}

        # Save VMCP_IMAGE for later steps
        echo "VMCP_IMAGE=${VMCP_IMAGE}" >> $GITHUB_ENV
        echo "Built and loaded vmcp image: ${VMCP_IMAGE}"

        # Pull and load all test server images in parallel to speed up CI
        echo "Pulling and loading test server images..."
        docker pull "$YARDSTICK_IMAGE" &
        docker pull ghcr.io/stackloklabs/gofetch/server:1.0.1 &
        docker pull ghcr.io/stackloklabs/osv-mcp/server:0.0.7 &
        docker pull python:3.9-slim &
        docker pull curlimages/curl:8.17.0 &
        docker pull ghcr.io/huggingface/text-embeddings-inference:cpu-latest &
        docker pull ghcr.io/dexidp/dex:v2.42.0 &
        wait

        # Load all images into kind
        kind load docker-image --name toolhive "$YARDSTICK_IMAGE"
        kind load docker-image --name toolhive ghcr.io/stackloklabs/gofetch/server:1.0.1
        kind load docker-image --name toolhive ghcr.io/stackloklabs/osv-mcp/server:0.0.7
        kind load docker-image --name toolhive python:3.9-slim
        kind load docker-image --name toolhive curlimages/curl:8.17.0
        kind load docker-image --name toolhive ghcr.io/huggingface/text-embeddings-inference:cpu-latest
        kind load docker-image --name toolhive ghcr.io/dexidp/dex:v2.42.0

    - name: Deploy operator with VMCP_IMAGE
      run: |
        export KUBECONFIG=kconfig.yaml
        echo "Deploying operator with vmcp image: ${VMCP_IMAGE}"

        # Build operator and proxyrunner images
        OPERATOR_IMAGE=$(KO_DOCKER_REPO=kind.local ko build --local -B ./cmd/thv-operator | tail -n 1)
        TOOLHIVE_IMAGE=$(KO_DOCKER_REPO=kind.local ko build --local -B ./cmd/thv-proxyrunner | tail -n 1)

        # Load operator images into kind
        kind load docker-image --name toolhive ${OPERATOR_IMAGE}
        kind load docker-image --name toolhive ${TOOLHIVE_IMAGE}

        # Deploy operator with VMCP_IMAGE environment variable
        helm upgrade --install toolhive-operator deploy/charts/operator \
          --set operator.image=${OPERATOR_IMAGE} \
          --set operator.toolhiveRunnerImage=${TOOLHIVE_IMAGE} \
          --set operator.vmcpImage=${VMCP_IMAGE} \
          --namespace toolhive-system \
          --create-namespace \
          --kubeconfig kconfig.yaml

        # Wait for operator to be ready
        kubectl rollout status deployment/toolhive-operator -n toolhive-system --timeout=2m --kubeconfig kconfig.yaml

    - name: Run VirtualMCP Lifecycle E2E tests
      run: |
        export KUBECONFIG=kconfig.yaml
        task thv-operator-e2e-test-run

    - name: Cleanup cluster
      if: always()
      run: |
        kind delete cluster --name toolhive