cost-model/.github/workflows/integration-testing.yaml

name: Run OpenCost Integration Tests

on:
    schedule:
      - cron: '0 14 * * *'
    push:
      branches:
        - develop
    pull_request:
      branches:
        - develop
    merge_group:
      types: [checks_requested]

concurrency:
    group: ${{ github.event.merge_group.head.sha || github.ref }}-intg-tests
    cancel-in-progress: false

jobs:
    check_actor_permissions:
        runs-on: ubuntu-latest
        permissions: {}
        outputs:
            ismaintainer: ${{ steps.check_permissions.outputs.ismaintainer }}
        steps:
          - name: Check if actor is a maintainer
            id: check_permissions
            uses: actions/github-script@v7
            with:
              script: |
                const { data: collaborators } = await github.rest.repos.listCollaborators({
                  owner: context.repo.owner,
                  repo: context.repo.repo,
                  permission: 'push'
                });
                
                const writers = collaborators.map(collaborator => collaborator.login);
                const isActorMaintainer = writers.includes(context.actor);
                
                console.log(`Actor: ${context.actor}`);
                console.log(`Repository writers: ${writers.join(', ')}`);
                console.log(`Is actor a maintainer? ${isActorMaintainer}`);
                
                core.setOutput('ismaintainer', isActorMaintainer.toString());
    noop-tests:
        needs: check_actor_permissions
        permissions: {}
        runs-on: Kubecost-Linux-Small-x86
        if: ${{ always() && !cancelled() && github.event_name != 'merge_group' && github.ref != 'refs/heads/develop' && needs.check_actor_permissions.outputs.ismaintainer == 'false' }}
        outputs:
            is_noop: ${{ steps.noop-tests.outputs.is_noop }}
        steps:
          - name: Tests Not Needed
            id: noop-tests
            run: |
              echo "integration tests not running because you are not a maintainer. they will run automatically when a PR is merged."
              echo "is_noop=true" >> $GITHUB_OUTPUT
    wait_for_image_ready:
        runs-on: ubuntu-latest
        permissions: {}
        needs: check_actor_permissions
        if: ${{ (always() && !cancelled()) && ( github.event_name == 'merge_group' || github.ref != 'refs/heads/develop'  || needs.check_actor_permissions.outputs.ismaintainer == 'true') }}
        outputs:
            IMAGE_TAG: ${{ steps.set_image_tags.outputs.IMAGE_TAG }}
            NAMESPACE: ${{ steps.set_image_tags.outputs.NAMESPACE }}
            MAINBRANCH: ${{ steps.set_image_tags.outputs.mainbranch }}
            passed: ${{ steps.wait_for_image_ready.outputs.passed }}
        steps:
          - uses: actions/checkout@v4
            with:
              ref: ${{ github.event.merge_group.head.sha || github.ref }}
          - name: Set OC SHA
            run: |
                 echo "OC_SHORTHASH=$(git rev-parse --short HEAD)" >> $GITHUB_ENV
          - name: Set image tags
            id: set_image_tags
            run: |
                    echo "github.event_name: ${{ github.event_name }}"
                    if [[ "${{ github.event_name }}" == "merge_group" ]]; then
                      echo "IMAGE_TAG=ghcr.io/${{ github.repository_owner }}/opencost:test-${{ steps.sha.outputs.OC_SHORTHASH }}" >> $GITHUB_OUTPUT
                      echo "NAMESPACE=merge-queue-${{ github.event.number }}-oc-${{ env.OC_SHORTHASH }}" >> $GITHUB_OUTPUT
                      echo "mainbranch=false" >> $GITHUB_OUTPUT
                    else
                      echo "building on develop branch"
                      echo "IMAGE_TAG=ghcr.io/${{ github.repository_owner }}/opencost:develop-${{ steps.sha.outputs.OC_SHORTHASH }}" >> $GITHUB_OUTPUT
                      echo "NAMESPACE=develop-oc-${{ env.OC_SHORTHASH }}" >> $GITHUB_OUTPUT
                      echo "mainbranch=true" >> $GITHUB_OUTPUT
                    fi

          - name: Log into ghcr.io
            uses: docker/login-action@v3
            with:
              registry: ghcr.io
              username: ${{ github.actor }}
              password: ${{ secrets.GITHUB_TOKEN }}

          - name: wait for docker image to be ready
            id: wait_for_image_ready
            run: |
                max_attempts=100
                # Loop until the Docker image can be pulled
                until docker manifest inspect ${{ steps.set_image_tags.outputs.IMAGE_TAG }}; do
                    echo "Waiting for Docker image ${{ steps.set_image_tags.outputs.IMAGE_TAG }} to be available, $max_attempts tries remain..."
                    sleep 6
                    max_attempts=$((max_attempts - 1))
                    if [[ $max_attempts -eq 0 ]]; then
                        echo "Docker image ${{ steps.set_image_tags.outputs.IMAGE_TAG }} is not available after 10 minutes. Exiting..."
                        exit 1
                    fi
                done

                echo "Docker image ${{ steps.set_image_tags.outputs.IMAGE_TAG }} is ready!"
                
                echo "passed=true" >> $GITHUB_OUTPUT
                
    build-test-stack:
        needs: wait_for_image_ready
        uses: opencost/opencost-infra/.github/workflows/build-stack.yaml@main
        secrets: inherit
        with:
            oc-container-version: "${{ needs.wait_for_image_ready.outputs.IMAGE_TAG }}"
            namespace: "${{ needs.wait_for_image_ready.outputs.NAMESPACE }}"

    wait-for-dns:
        needs: [wait_for_image_ready, build-test-stack]
        runs-on: ubuntu-latest
        permissions: {}
        steps:
          - name: Wait for DNS to resolve
            run: |
              echo "Waiting for ${{ needs.wait_for_image_ready.outputs.NAMESPACE }}.infra.opencost.io to resolve in DNS..."
              
              max_attempts=60
              until host ${{ needs.wait_for_image_ready.outputs.NAMESPACE }}.infra.opencost.io; do
                echo "DNS not yet resolved for ${{ needs.wait_for_image_ready.outputs.NAMESPACE }}.infra.opencost.io, $max_attempts tries remain..."
                sleep 10
                max_attempts=$((max_attempts - 1))
                if [[ $max_attempts -eq 0 ]]; then
                  echo "DNS resolution failed for ${{ needs.wait_for_image_ready.outputs.NAMESPACE }}.infra.opencost.io after 10 minutes. Exiting..."
                  exit 1
                fi
              done
              
              echo "DNS resolved successfully for ${{ needs.wait_for_image_ready.outputs.NAMESPACE }}.infra.opencost.io!"
    run-tests:
        needs: [wait_for_image_ready, build-test-stack, wait-for-dns]
        permissions: {}
        uses: opencost/opencost-infra/.github/workflows/test-stack.yaml@main
        secrets: inherit
        with:
            namespace: "${{ needs.wait_for_image_ready.outputs.NAMESPACE }}"
            target_branch: "${{ github.event.pull_request.head.ref || 'main' }}"
    
    teardown-test-stack:
        needs: [wait_for_image_ready, run-tests]
        permissions: {}
        uses: opencost/opencost-infra/.github/workflows/destroy-stack.yaml@main
        if: always()
        secrets: inherit 
        with:
            namespace: "${{ needs.wait_for_image_ready.outputs.NAMESPACE }}"

    check-success:
        needs: [noop-tests, run-tests]
        permissions: {}
        runs-on: ubuntu-latest
        if: ${{ always() }}
        steps:
          - name: Check success
            run: |
              if [[ "${{ needs.noop-tests.outputs.is_noop }}" == "true" ]]; then
                echo "No-op tests, skipping success check"
                exit 0
              fi
              
              if [[ "${{ needs.run-tests.outputs.passed }}" != "true" ]]; then
                echo "One or more integration tests failed"
                exit 1
              fi
              
              echo "All integration tests passed"
              exit 0

    set-labels:
      needs: [wait_for_image_ready, run-tests]
      runs-on: ubuntu-latest
      permissions: {}
      steps:
        - name: label integration tests failing
          if: ${{ always()  && contains(needs.*.result, 'failure') && !cancelled()}}
          uses: andymckay/labeler@1.0.4
          with:
            add-labels: "integration tests failed"
        - uses: mondeja/remove-labels-gh-action@v2
          if: ${{ always()  && contains(needs.*.result, 'failure') && !cancelled()}}
          with:
                token: ${{ secrets.GITHUB_TOKEN }}
                labels: |
                  integration tests passed
        - name: Label integration tests passing
          if: ${{ always()  && !contains(needs.*.result, 'failure') && !cancelled()}}
          uses: andymckay/labeler@1.0.4
          with:
            add-labels: "integration tests passed"
        - uses: mondeja/remove-labels-gh-action@v2
          if: ${{ always()  && !contains(needs.*.result, 'failure') && !cancelled()}}
          with:
              token: ${{ secrets.GITHUB_TOKEN }}
              labels: |
                integration tests failed