GenAIExamples/.github/workflows/_manifest-e2e.yml

# Copyright (C) 2024 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

name: Single Kubernetes Manifest E2e Test For Call
on:
  workflow_call:
    inputs:
      example:
        default: "ChatQnA"
        description: "The example to test on K8s"
        required: true
        type: string
      hardware:
        default: "xeon"
        description: "Nodes to run the test, xeon or gaudi"
        required: true
        type: string
      tag:
        default: "latest"
        description: "Tag to apply to images, default is latest"
        required: false
        type: string

jobs:
  manifest-test:
    runs-on: "k8s-${{ inputs.hardware }}"
    continue-on-error: true
    steps:
      - name: Clean Up Working Directory
        run: sudo rm -rf ${{github.workspace}}/*

      - name: Get checkout ref
        run: |
          if [ "${{ github.event_name }}" == "pull_request" ] || [ "${{ github.event_name }}" == "pull_request_target" ]; then
            echo "CHECKOUT_REF=refs/pull/${{ github.event.number }}/merge" >> $GITHUB_ENV
          else
            echo "CHECKOUT_REF=${{ github.ref }}" >> $GITHUB_ENV
          fi
          echo "checkout ref ${{ env.CHECKOUT_REF }}"

      - name: Checkout out Repo
        uses: actions/checkout@v4
        with:
          ref: ${{ env.CHECKOUT_REF }}
          fetch-depth: 0

      - name: Set variables
        run: |
          echo "IMAGE_REPO=${OPEA_IMAGE_REPO}opea" >> $GITHUB_ENV
          echo "IMAGE_TAG=${{ inputs.tag }}" >> $GITHUB_ENV
          lower_example=$(echo "${{ inputs.example }}" | tr '[:upper:]' '[:lower:]')
          echo "NAMESPACE=$lower_example-$(tr -dc a-z0-9 </dev/urandom | head -c 16)" >> $GITHUB_ENV
          echo "ROLLOUT_TIMEOUT_SECONDS=1800s" >> $GITHUB_ENV
          echo "KUBECTL_TIMEOUT_SECONDS=60s" >> $GITHUB_ENV
          echo "continue_test=true" >> $GITHUB_ENV
          echo "should_cleanup=false" >> $GITHUB_ENV
          echo "skip_validate=true" >> $GITHUB_ENV
          echo "NAMESPACE=$NAMESPACE"

      - name: Kubectl install
        id: install
        run: |
          if [[ ! -f ${{ github.workspace }}/${{ inputs.example }}/tests/test_manifest_on_${{ inputs.hardware }}.sh ]]; then
            echo "No test script found, exist test!"
            exit 0
          else
            ${{ github.workspace }}/${{ inputs.example }}/tests/test_manifest_on_${{ inputs.hardware }}.sh init_${{ inputs.example }}
            echo "should_cleanup=true" >> $GITHUB_ENV
            kubectl create ns $NAMESPACE
            ${{ github.workspace }}/${{ inputs.example }}/tests/test_manifest_on_${{ inputs.hardware }}.sh install_${{ inputs.example }} $NAMESPACE
            echo "Testing ${{ inputs.example }}, waiting for pod ready..."
            if kubectl rollout status deployment --namespace "$NAMESPACE" --timeout "$ROLLOUT_TIMEOUT_SECONDS"; then
              echo "Testing manifests ${{ inputs.example }}, waiting for pod ready done!"
              echo "skip_validate=false" >> $GITHUB_ENV
            else
              echo "Timeout waiting for pods in namespace $NAMESPACE to be ready!"
              .github/workflows/scripts/k8s-utils.sh dump_pods_status $NAMESPACE
              exit 1
            fi
            sleep 60
          fi

      - name: Validate e2e test
        if: always()
        run: |
          if $skip_validate; then
            echo "Skip validate"
          else
            if ${{ github.workspace }}/${{ inputs.example }}/tests/test_manifest_on_${{ inputs.hardware }}.sh validate_${{ inputs.example }} $NAMESPACE ; then
              echo "Validate ${{ inputs.example }} successful!"
            else
              echo "Validate ${{ inputs.example }} failure!!!"
              echo "Check the logs in 'Dump logs when e2e test failed' step!!!"
              exit 1
            fi
          fi

      - name: Dump logs when e2e test failed
        if: failure()
        run: |
          .github/workflows/scripts/k8s-utils.sh dump_all_pod_logs $NAMESPACE

      - name: Kubectl uninstall
        if: always()
        run: |
          if $should_cleanup; then
            if ! kubectl delete ns $NAMESPACE --timeout=$KUBECTL_TIMEOUT_SECONDS; then
              kubectl delete pods --namespace $NAMESPACE --force --grace-period=0 --all
              kubectl delete ns $NAMESPACE --force --grace-period=0 --timeout=$KUBECTL_TIMEOUT_SECONDS
            fi
          fi