e2e-test.yaml

name: e2e-test

on:
  push:
    branches:
      - main
  pull_request:

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

env:
  REPOSITORY: koperator_e2e_test

jobs:
  build:
    runs-on: ubuntu-26.04

    steps:
      - name: Checkout code
        uses: actions/checkout@v7

      - name: Clean workspace
        run: |
          # Remove any existing go.work files to ensure clean module resolution
          rm -f go.work go.work.sum

      - name: Set up Go
        uses: actions/setup-go@v7
        with:
          go-version-file: go.mod

      - name: Setup Helm
        uses: azure/setup-helm@v5
        with:
          version: '4.3.0' # renovate: datasource=github-releases depName=helm/helm

      - name: Clean Go module cache
        run: |
          # Clean any existing Go module cache to avoid conflicts
          # Use sudo to handle permission issues and force removal
          sudo rm -rf ~/go/pkg/mod || true
          sudo rm -rf ~/.cache/go-build || true
          mkdir -p ~/go/pkg/mod
          mkdir -p ~/.cache/go-build
          # Ensure proper permissions
          chmod -R 755 ~/go/pkg/mod
          chmod -R 755 ~/.cache/go-build

      # Enable Tmate Session to debug the E2E Kind Cluster
      # Only activates when re-running the workflow in debug mode
      - name: Setup tmate session
        if: ${{ runner.debug == '1' }}
        continue-on-error: true
        uses: mxschmitt/action-tmate@v3
        with:
          detached: true
          limit-access-to-actor: true

      - name: Build docker image
        run: |
          IMG=$REPOSITORY:$GITHUB_SHA make docker-build

      - name: Setup Kind cluster
        id: setup-kind
        uses: ./.github/actions/kind-create

      - name: Load image into kind cluster
        run: |
          kind load docker-image $REPOSITORY:$GITHUB_SHA --name e2e-kind


      - name: Download dependencies
        run: |
          # Download dependencies for all modules
          go mod download
          cd api && go mod download && cd ..
          cd properties && go mod download && cd ..
          cd tests/e2e && go mod download && cd ../..


      - name: Run E2E tests
        env:
           KUBECONFIG: ${{ steps.setup-kind.outputs.kubeconfig }}
        run: |
          # Clean any existing go.work files to avoid workspace conflicts
          rm -f go.work go.work.sum
          IMG_E2E=$REPOSITORY:$GITHUB_SHA make test-e2e

      # Dump cluster diagnostics on failure so e2e failures (e.g. a dependency Helm
      # install that times out under `--atomic`) are debuggable from the job log
      # without needing a live tmate session.
      - name: Dump diagnostics on failure
        if: failure()
        env:
          KUBECONFIG: ${{ steps.setup-kind.outputs.kubeconfig }}
        run: |
          set +e
          echo "::group::nodes"
          kubectl get nodes -o wide
          echo "::endgroup::"
          echo "::group::all pods"
          kubectl get pods -A -o wide
          echo "::endgroup::"
          for ns in cert-manager projectcontour zookeeper prometheus kafka; do
            echo "::group::namespace ${ns}"
            kubectl get pods -n "${ns}" -o wide
            echo "----- events (${ns}) -----"
            kubectl get events -n "${ns}" --sort-by=.lastTimestamp | tail -50
            echo "----- describe pods (${ns}) -----"
            kubectl describe pods -n "${ns}"
            echo "----- container logs (${ns}) -----"
            for pod in $(kubectl get pods -n "${ns}" -o name); do
              echo ">>> ${pod} (current) <<<"
              kubectl logs -n "${ns}" "${pod}" --all-containers --tail=150 --timestamps
              echo ">>> ${pod} (previous, if the container restarted) <<<"
              kubectl logs -n "${ns}" "${pod}" --all-containers --previous --tail=150 --timestamps
            done
            echo "::endgroup::"
          done
          echo "::group::warning events (all namespaces)"
          kubectl get events -A --field-selector type=Warning --sort-by=.lastTimestamp | tail -80
          echo "::endgroup::"
          # Broker/disk removal stalls are almost always a Cruise Control problem: a stuck CC
          # Deployment rollout (two revisions racing) or a remove_broker task wedged InExecution.
          # None of that is visible in the per-namespace pod dump above, so capture the koperator
          # CRs and the CC rollout state explicitly.
          echo "::group::kafka CRs + cruisecontrol rollout"
          kubectl -n kafka get kafkacluster,cruisecontroloperation -o wide
          echo "----- kafkacluster status.state -----"
          kubectl -n kafka get kafkacluster -o jsonpath='{range .items[*]}{.metadata.name}{": "}{.status.state}{"\n"}{end}'
          echo "----- cruisecontroloperation (yaml: currentTask.state / errorMessage) -----"
          kubectl -n kafka get cruisecontroloperation -o yaml
          echo "----- cruisecontrol deployment + replicasets (exposes stuck/duplicate revisions) -----"
          kubectl -n kafka get deploy,rs -l app=cruisecontrol -o wide
          kubectl -n kafka describe deploy -l app=cruisecontrol
          echo "::endgroup::"
          # Disk-removal stalls can also leave orphaned PVCs behind: a broker's old volume
          # gets deleted once Cruise Control confirms removal, but the delete only completes
          # once the pvc-protection finalizer clears (i.e. nothing still mounts it). `get -o wide`
          # surfaces a stuck `Terminating` STATUS at a glance; `describe` shows the finalizers/
          # deletionTimestamp and any FailedAttachVolume/VolumeFailedDelete events explaining why.
          echo "::group::persistentvolumeclaims (kafka)"
          kubectl -n kafka get pvc -o wide
          echo "----- describe pvc (kafka) -----"
          kubectl -n kafka describe pvc
          echo "::endgroup::"
          exit 0