flake-validation.yaml

name: Flake Validation

# Run a single E2E test suite N times in parallel (no gotestsum retries) to
# confirm a flake fix is solid before merging. Trigger manually and point it at
# whichever suite is under scrutiny.

on:
  workflow_dispatch:
    inputs:
      run-regex:
        description: "go -run regex for the test suite to validate (e.g. '^TestKgateway$$/^Backends$$')"
        required: true
        type: string
      repetitions:
        description: "Number of parallel cluster runs"
        required: false
        default: '5'
        type: choice
        options:
          - '3'
          - '5'
          - '10'

env:
  VERSION: 'v1.0.0-ci1'
  GITHUB_TOKEN: ${{ github.token }}
  # Intentionally 0: retries hide flakes; we want raw pass/fail.
  GO_TEST_RETRIES: 0
  ENVOYINIT_CACHE_REF: ghcr.io/${{ github.repository_owner }}/envoy-wrapper-cache
  CONTROLLER_CACHE_REF: ghcr.io/${{ github.repository_owner }}/kgateway-cache
  SDS_CACHE_REF: ghcr.io/${{ github.repository_owner }}/sds-cache

jobs:
  build_e2e_images:
    name: Build End-to-End Images
    runs-on: ubuntu-22.04
    timeout-minutes: 45
    steps:
      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
      - name: Prep Go Runner
        uses: ./.github/actions/prep-go-runner
      - name: Setup Buildx
        uses: docker/setup-buildx-action@b5ca514318bd6ebac0fb2aedd5d36ec1b5c232a2 # v3.10.0
      - name: Restore envoyinit BuildKit cache
        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
        with:
          path: /tmp/.buildx-cache-envoyinit
          key: ${{ runner.os }}-buildx-envoyinit-v1-${{ hashFiles('internal/envoy_modules/Cargo.lock', 'internal/envoy_modules/**/Cargo.toml', 'cmd/envoyinit/Dockerfile.tmpl', 'cmd/envoyinit/generate-dockerfile.sh') }}
          restore-keys: |
            ${{ runner.os }}-buildx-envoyinit-v1-
      - name: Build shared e2e images
        env:
          ENVOYINIT_LOCAL_CACHE_FROM: /tmp/.buildx-cache-envoyinit
          ENVOYINIT_LOCAL_CACHE_TO: /tmp/.buildx-cache-envoyinit-new
        run: |
          mkdir -p /tmp/.buildx-cache-envoyinit
          make save-e2e-shared-images
      - name: Rotate envoyinit BuildKit cache
        if: always()
        run: |
          if [ -d /tmp/.buildx-cache-envoyinit-new ]; then
            rm -rf /tmp/.buildx-cache-envoyinit
            mv /tmp/.buildx-cache-envoyinit-new /tmp/.buildx-cache-envoyinit
          fi
      - name: Upload shared e2e images
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: e2e-shared-images
          path: _output/e2e-images/shared-images.tar
          if-no-files-found: error
          retention-days: 1
          compression-level: 0

  generate-matrix:
    name: Generate repetition matrix
    runs-on: ubuntu-22.04
    outputs:
      matrix: ${{ steps.matrix.outputs.matrix }}
    steps:
      - id: matrix
        shell: bash
        run: |
          COUNT=${{ inputs.repetitions }}
          echo "matrix=$(seq 1 "$COUNT" | jq -cs '.')" >> "$GITHUB_OUTPUT"

  flake_validation:
    name: Flake Validation (run ${{ matrix.run-number }})
    needs: [build_e2e_images, generate-matrix]
    runs-on: ubuntu-22.04
    timeout-minutes: 60
    strategy:
      fail-fast: false
      matrix:
        run-number: ${{ fromJson(needs.generate-matrix.outputs.matrix) }}
        version-files:
          - file: './.github/workflows/.env/pr-tests/versions.env'
    steps:
      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
      - name: Prep Go Runner
        uses: ./.github/actions/prep-go-runner
      - name: Dotenv Action
        uses: falti/dotenv-action@a33be0b8cf6a6e6f1b82cc9f3782061ab1022be5 # v1.1.4
        id: dotenv
        with:
          path: ${{ matrix.version-files.file }}
          log-variables: true
      - name: Download shared e2e images
        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        with:
          name: e2e-shared-images
          path: _output/e2e-images
      - name: Load shared e2e images
        shell: bash
        run: docker load --input _output/e2e-images/shared-images.tar
      - id: setup-kind-cluster
        name: Setup KinD Cluster
        if: ${{ steps.dotenv.outputs.cluster_type != 'k3d' }}
        uses: ./.github/actions/setup-kind-cluster
        env:
          SKIP_DOCKER: "true"
          LOAD_DOCKER_IMAGES: "true"
        with:
          cluster-name: flake-${{ matrix.run-number }}
          kind-node-version: ${{ steps.dotenv.outputs.node_version }}
          kubectl-version: ${{ steps.dotenv.outputs.kubectl_version }}
          istio-version: ${{ steps.dotenv.outputs.istio_version }}
          localstack: 'false'
      - id: setup-k3d-cluster
        name: Setup k3d Cluster
        if: ${{ steps.dotenv.outputs.cluster_type == 'k3d' }}
        uses: ./.github/actions/setup-k3d-cluster
        env:
          SKIP_DOCKER: "true"
          LOAD_DOCKER_IMAGES: "true"
        with:
          cluster-name: flake-${{ matrix.run-number }}
          node-version: ${{ steps.dotenv.outputs.node_version }}
          kubectl-version: ${{ steps.dotenv.outputs.kubectl_version }}
          istio-version: ${{ steps.dotenv.outputs.istio_version }}
          localstack: 'false'
          github-token: ${{ github.token }}
      - name: Load shared extproc server
        shell: bash
        env:
          CLUSTER_NAME: flake-${{ matrix.run-number }}
          CLUSTER_TYPE: ${{ steps.dotenv.outputs.cluster_type || 'kind' }}
        run: make cluster-load-extproc-server
      - id: run-tests
        uses: ./.github/actions/kubernetes-e2e-tests
        env:
          SKIP_EXTPROC_SERVER_SETUP: "true"
        with:
          cluster-name: flake-${{ matrix.run-number }}
          cluster-type: ${{ steps.dotenv.outputs.cluster_type || 'kind' }}
          test-args: '-timeout=25m'
          run-regex: ${{ inputs.run-regex }}
          istio-version: ${{ steps.dotenv.outputs.istio_version }}
          matrix-label: flake-${{ matrix.run-number }}