containerd/.github/workflows/windows-periodic.yml

# Workflow intended to run containerd integration tests on Windows.

name: Windows Integration Tests

on:
  workflow_dispatch:
  workflow_call:
    secrets:
      AZURE_SUB_ID:
        required: true
      AZURE_CREDS:
        required: true
      GCP_SERVICE_ACCOUNT:
        required: true
      GCP_WORKLOAD_IDENTITY_PROVIDER:
        required: true

env:
  AZURE_DEFAULT_LOCATION: westeurope
  AZURE_SUBSCRIPTION_ID: ${{ secrets.AZURE_SUB_ID }}
  AZURE_DEFAULT_VM_SIZE: Standard_D2s_v3
  PASSWORD: Passw0rdAdmin # temp for testing, will be generated
  DEFAULT_ADMIN_USERNAME: azureuser
  SSH_OPTS: "-o ServerAliveInterval=20 -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null"
  REMOTE_VM_BIN_PATH: "c:\\containerd\\bin"
  BUSYBOX_TESTING_IMAGE_REF: "k8s.gcr.io/e2e-test-images/busybox:1.29-2"
  RESOURCE_CONSUMER_TESTING_IMAGE_REF: "k8s.gcr.io/e2e-test-images/resource-consumer:1.10"
  WEBSERVER_TESTING_IMAGE_REF: "k8s.gcr.io/e2e-test-images/nginx:1.14-2"


jobs:
  winIntegration:
    # NOTE: the following permissions are required by `google-github-actions/auth`:
    permissions:
      contents: 'read'
      id-token: 'write'
    strategy:
      # NOTE(aznashwan): this will permit all other jobs from the matrix to finish and
      # upload their results even if one has a failing non-test-task:
      # (e.g. hitting resource limits in the `AZTestVMCreate` task)
      fail-fast: false
      matrix:
        win_ver: [ltsc2019, ltsc2022]
        include:
        - win_ver: ltsc2019
          AZURE_IMG: "MicrosoftWindowsServer:WindowsServer:2019-Datacenter-with-Containers-smalldisk:17763.2565.220202"
          AZURE_RESOURCE_GROUP: ctrd-integration-ltsc2019-${{ github.run_id }}
          GOOGLE_BUCKET: "containerd-integration/logs/windows-ltsc2019/"
        - win_ver: ltsc2022
          AZURE_IMG: "MicrosoftWindowsServer:WindowsServer:2022-datacenter-smalldisk-g2:20348.524.220201"
          AZURE_RESOURCE_GROUP: ctrd-integration-ltsc2022-${{ github.run_id }}
          GOOGLE_BUCKET: "containerd-integration/logs/windows-ltsc2022/"
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2

      - name: Install required packages
        run: |
          sudo apt-get install xmlstarlet -y

      - name: PrepareArtifacts
        run: |
          STARTED_TIME=$(date +%s)
          LOGS_DIR=$HOME/$STARTED_TIME
          echo "STARTED_TIME=$STARTED_TIME" >> $GITHUB_ENV
          echo "LOGS_DIR=$LOGS_DIR" >> $GITHUB_ENV

          echo "VM_INTEGRATION_LOGFILE=/c/Logs/integration.log" >> $GITHUB_ENV
          echo "VM_CRI_INTEGRATION_LOGFILE=/c/Logs/cri-integration.log" >> $GITHUB_ENV

          mkdir -p $LOGS_DIR/artifacts
          jq -n --arg node temp --arg timestamp $STARTED_TIME '$timestamp|tonumber|{timestamp:.,$node}' > $LOGS_DIR/started.json

      - name: Generate ssh key pair
        run: |
          mkdir -p $HOME/.ssh/
          ssh-keygen -t rsa -b 4096 -C "ci@containerd.com" -f $HOME/.ssh/id_rsa -q -N ""
          echo "SSH_PUB_KEY=$(cat ~/.ssh/id_rsa.pub)" >> $GITHUB_ENV

      - name: AZLogin
        uses: azure/login@v1
        with:
          creds: ${{ secrets.AZURE_CREDS }}

      - name: AZResourceGroupCreate
        uses: azure/CLI@v1
        with:
          inlinescript: |
            az group create -n ${{ matrix.AZURE_RESOURCE_GROUP }} -l ${{ env.AZURE_DEFAULT_LOCATION }} --tags creationTimestamp=$(date -u '+%Y-%m-%dT%H:%M:%SZ')

      - name: AZTestVMCreate
        uses: azure/CLI@v1
        with:
          inlinescript: |
            DETAILS=$(az vm create -n winTestVM --admin-username ${{ env.DEFAULT_ADMIN_USERNAME }} --admin-password ${{ env.PASSWORD }} --image ${{ matrix.AZURE_IMG }} -g ${{ matrix.AZURE_RESOURCE_GROUP }} --nsg-rule SSH --size ${{ env.AZURE_DEFAULT_VM_SIZE }} --public-ip-sku Standard -o json)
            PUB_IP=$(echo $DETAILS | jq -r .publicIpAddress)
            if [ "$PUB_IP" == "null" ]
            then
                RETRY=0
                while [ "$PUB_IP" == "null" ] || [ $RETRY -le 5 ]
                do
                    sleep 5
                    PUB_IP=$(az vm show -d -g ${{ matrix.AZURE_RESOURCE_GROUP }} -n winTestVM -o json --query publicIps | jq -r)
                    RETRY=$(( $RETRY + 1 ))
                done
            fi

            if [ "$PUB_IP" == "null" ]
            then
                echo "failed to fetch public IP"
                exit 1
            fi
            echo "VM_PUB_IP=$PUB_IP" >> $GITHUB_ENV

      - name: EnableAZVMSSH
        uses: azure/CLI@v1
        with:
          inlinescript: |
            az vm run-command invoke  --command-id RunPowerShellScript -n winTestVM -g ${{ matrix.AZURE_RESOURCE_GROUP }} --scripts @$GITHUB_WORKSPACE/script/setup/enable_ssh_windows.ps1 --parameters 'SSHPublicKey=${{ env.SSH_PUB_KEY }}'

      - name: TestSSHConnection
        run: |
          if ! ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }}  "hostname";
          then
            exit 1
          fi

      - name: InstallContainerFeatureWS2022
        if: ${{ matrix.win_ver == 'ltsc2022' }}
        run: |
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "powershell.exe -command { Install-WindowsFeature -Name 'Containers' -Restart }"

      - name: WaitForVMToRestart
        if: ${{ matrix.win_ver == 'ltsc2022' }}
        timeout-minutes: 5
        run: |
          # give the vm 30 seconds to actually stop. SSH server might actually respond while server is shutting down.
          sleep 30
          while [ ! $( ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }}  "hostname") ];
          do
             echo "Unable to connect to azurevm"
          done
          echo "Connection reestablished. VM restarted succesfully."

      - name: CreateNatNetworkWS2022
        if: ${{ matrix.win_ver == 'ltsc2022' }}
        run: |
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "powershell.exe -command { curl.exe -L 'https://raw.githubusercontent.com/microsoft/SDN/master/Kubernetes/windows/hns.psm1' -o hns.psm1 }"
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "powershell.exe -command { Import-Module .\hns.psm1 ; New-HnsNetwork -Type NAT -Name nat -AddressPrefix 172.19.208.0/20 -Gateway 172.19.208.1 }"

      - name: PrepareTestingEnv
        run: |
          scp -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }}  $GITHUB_WORKSPACE/script/setup/prepare_env_windows.ps1  azureuser@${{ env.VM_PUB_IP }}:/prepare_env_windows.ps1
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }}  azureuser@${{ env.VM_PUB_IP }} "c:\\prepare_env_windows.ps1"

      - name: MakeContainerDBins
        run: |
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "git clone http://github.com/containerd/containerd c:\\containerd "
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "cd c:\containerd ; make binaries"

      - name: RunIntegrationTests
        id: RunIntegrationTests
        # NOTE(aznashwan): this is set to continue-on-error to allow the workflow to run until
        # the reports are converted/uploaded to GCloud so as to show up on testgrid.k8s.io too.
        continue-on-error: true
        run: |
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "sh.exe -s" << EOF
            cd /c/containerd
            export EXTRA_TESTFLAGS="-timeout=20m"
            set -o pipefail
            make integration | tee ${{ env.VM_INTEGRATION_LOGFILE }}
          EOF
          echo '::set-output name=SUCCEEDED::1'

      - name: PrepareRepoList
        run: |
            cat > repolist.toml << EOF
            busybox = "${{ env.BUSYBOX_TESTING_IMAGE_REF }}"
            ResourceConsumer = "${{ env.RESOURCE_CONSUMER_TESTING_IMAGE_REF }}"
            EOF

            cat > cri-test-images.yaml << EOF
            defaultTestContainerImage: ${{ env.BUSYBOX_TESTING_IMAGE_REF }}
            webServerTestImage: ${{ env.WEBSERVER_TESTING_IMAGE_REF }}
            EOF

            scp -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} repolist.toml azureuser@${{ env.VM_PUB_IP }}:c:/repolist.toml
            scp -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} cri-test-images.yaml azureuser@${{ env.VM_PUB_IP }}:c:/cri-test-images.yaml

      - name: RunCRIIntegrationTests
        id: RunCRIIntegrationTests
        # NOTE(aznashwan): this is set to continue-on-error to allow the workflow to run until
        # the reports are converted/uploaded to GCloud so as to show up on testgrid.k8s.io too.
        continue-on-error: true
        run: |
           ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "sh.exe -s" <<EOF
             cd c:/containerd
             ./script/setup/install-cni-windows
             export TEST_IMAGE_LIST=c:/repolist.toml
             set -o pipefail
             make cri-integration | tee ${{ env.VM_CRI_INTEGRATION_LOGFILE }}
           EOF
           echo '::set-output name=SUCCEEDED::1'

      - name: GetCritestRepo
        run: |
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "git clone https://github.com/kubernetes-sigs/cri-tools c:/cri-tools"

      - name: BuildCritest
        run: |
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "sh.exe -c 'cd /c/cri-tools && make critest'"

      - name: RunCritest
        id: RunCritest
        # NOTE(aznashwan): this is set to continue-on-error to allow the workflow to run until
        # the reports are converted/uploaded to GCloud so as to show up on testgrid.k8s.io too.
        continue-on-error: true
        run: |
          # This test is exceedingly flaky only on ws2022 so skip for now to keep CI happy.
          # Info: https://github.com/containerd/containerd/issues/6652
          SKIP=""
          if [ '${{ matrix.win_ver }}' == 'ltsc2022' ];then
            SKIP='-ginkgo.skip="runtime should support exec with tty=true and stdin=true"'
          fi

          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "powershell.exe -command { C:\containerd\bin\containerd.exe --log-level=debug --log-file=C:/logs/containerd.log --service-name containerd --register-service ; Set-Service containerd -StartupType Automatic; Start-Service containerd }"
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "sh.exe -s" <<EOF
            sleep 5
            set -o pipefail
            c:/cri-tools/build/bin/critest.exe $SKIP --runtime-endpoint='npipe://./pipe/containerd-containerd' --test-images-file='c:/cri-test-images.yaml' --report-dir='c:/Logs' -ginkgo.junit-report="C:\Logs\junit_critest.xml" | tee c:/Logs/critest.log
          EOF
          echo '::set-output name=SUCCEEDED::1'

      - name: PullLogsFromWinNode
        run: |
          # Generate JUnit reports from the stdouts of the tests:
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "sh.exe -c 'touch ${{ env.VM_INTEGRATION_LOGFILE }}; cat ${{ env.VM_INTEGRATION_LOGFILE }} | go-junit-report.exe > /c/Logs/junit_integration.xml'"
          ssh -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }} "sh.exe -c 'touch ${{ env.VM_CRI_INTEGRATION_LOGFILE }}; cat ${{ env.VM_CRI_INTEGRATION_LOGFILE }} | go-junit-report.exe > /c/Logs/junit_cri_integration.xml'"

          # Copy over all the JUnit reports:
          scp -i $HOME/.ssh/id_rsa ${{ env.SSH_OPTS }} azureuser@${{ env.VM_PUB_IP }}:c:/Logs/*.xml ${{ env.LOGS_DIR }}/artifacts/
          for f in $(ls ${{ env.LOGS_DIR }}/artifacts/*.xml); do
              xmlstarlet ed -d "/testsuites/testsuite/properties" $f > ${{ env.LOGS_DIR }}/$(basename $f)
              mv ${{ env.LOGS_DIR }}/$(basename $f) $f
          done

      - name: FinishJob
        run: |
          jq -n --arg result SUCCESS --arg timestamp $(date +%s) '$timestamp|tonumber|{timestamp:.,$result}' > ${{ env.LOGS_DIR }}/finished.json
          echo "${{ env.STARTED_TIME }}" > ${{ github.workspace }}/latest-build.txt

      - name: AssignGcpCreds
        id: AssignGcpCreds
        run: |
          echo '::set-output name=GCP_SERVICE_ACCOUNT::${{ secrets.GCP_SERVICE_ACCOUNT }}'
          echo '::set-output name=GCP_WORKLOAD_IDENTITY_PROVIDER::${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}'

      - name: AuthGcp
        uses: google-github-actions/auth@v0
        if: steps.AssignGcpCreds.outputs.GCP_SERVICE_ACCOUNT && steps.AssignGcpCreds.outputs.GCP_WORKLOAD_IDENTITY_PROVIDER
        with:
          service_account: ${{ secrets.GCP_SERVICE_ACCOUNT }}
          workload_identity_provider: ${{ secrets.GCP_WORKLOAD_IDENTITY_PROVIDER }}

      - name: UploadJobReport
        uses: google-github-actions/upload-cloud-storage@v0.8.0
        if: steps.AssignGcpCreds.outputs.GCP_SERVICE_ACCOUNT && steps.AssignGcpCreds.outputs.GCP_WORKLOAD_IDENTITY_PROVIDER
        with:
          path: ${{ github.workspace }}/latest-build.txt
          destination: ${{ matrix.GOOGLE_BUCKET }}
          parent: false

      - name: UploadLogsDir
        uses: google-github-actions/upload-cloud-storage@v0.8.0
        if: steps.AssignGcpCreds.outputs.GCP_SERVICE_ACCOUNT && steps.AssignGcpCreds.outputs.GCP_WORKLOAD_IDENTITY_PROVIDER
        with:
          path: ${{ env.LOGS_DIR }}
          destination: ${{ matrix.GOOGLE_BUCKET }}${{ env.STARTED_TIME}}
          parent: false

      - name: Check all CI stages succeeded
        uses: actions/github-script@v3
        with:
          script: |
            const stepResults = {
              RunIntegrationTests: "${{ steps.RunIntegrationTests.outputs.SUCCEEDED }}",
              RunCRIIntegrationTests: "${{ steps.RunCRIIntegrationTests.outputs.SUCCEEDED }}",
              RunCritest: "${{ steps.RunCritest.outputs.SUCCEEDED }}",
            };

            let failedTasks = [];
            for( [step, result] of Object.entries(stepResults) ) {
              if (result != "1") {
                failedTasks.push(step);
              }
            };

            if (failedTasks.length != 0) {
              core.setFailed(`One or more CI stages have failed. Please review the outputs of the following stepts: ${failedTasks}.`);
            };

      - name: ResourceCleanup
        if: always()
        uses: azure/CLI@v1
        with:
          inlinescript: |
            az group delete -g ${{ matrix.AZURE_RESOURCE_GROUP }} --yes