docs/.github/workflows/azure-prod-build-deploy.yml

name: Azure Production - Build and Deploy

# **What it does**: Builds and deploys the default branch to production
# **Why we have it**: To enable us to deploy the latest to production whenever necessary rather than relying on PR merges.
# **Who does it impact**: All contributors.

on:
  push:
    branches:
      - main
  workflow_dispatch:

permissions:
  contents: read
  deployments: write

# This allows a subsequently queued workflow run to take priority over
# previously queued runs but NOT interrupt currently executing runs
concurrency:
  group: '${{ github.workflow }}'
  cancel-in-progress: false

jobs:
  azure-prod-build-and-deploy:
    if: ${{ github.repository == 'github/docs-internal' }}
    runs-on: ubuntu-20.04-xl
    timeout-minutes: 20
    environment:
      name: production
      url: 'https://docs.github.com'
    env:
      DOCKER_IMAGE: ${{ secrets.PROD_REGISTRY_SERVER }}/${{ github.repository }}:${{ github.sha }}
      DOCKER_IMAGE_CACHE_REF: ${{ secrets.PROD_REGISTRY_SERVER }}/${{ github.repository }}:main-production

    steps:
      - name: 'Az CLI login'
        uses: azure/login@92a5484dfaf04ca78a94597f4f19fea633851fa2 # pin @v1.4.6
        with:
          creds: ${{ secrets.PROD_AZURE_CREDENTIALS }}

      - name: 'Docker login'
        uses: azure/docker-login@83efeb77770c98b620c73055fbb59b2847e17dc0
        with:
          login-server: ${{ secrets.PROD_REGISTRY_SERVER }}
          username: ${{ secrets.PROD_REGISTRY_USERNAME }}
          password: ${{ secrets.PROD_REGISTRY_PASSWORD }}

      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@95cb08cb2672c73d4ffd2f422e6d11953d2a9c70

      - name: Check out repo
        uses: actions/checkout@93ea575cb5d8a053eaa0ac8fa3b40d7e05a33cc8
        with:
          ref: ${{ github.sha }}
          # To prevent issues with cloning early access content later
          persist-credentials: 'false'

      - name: Setup Node.js
        uses: actions/setup-node@8c91899e586c5b171469028077307d293428b516
        with:
          node-version-file: 'package.json'
          cache: npm

      - name: Clone docs-early-access
        uses: actions/checkout@93ea575cb5d8a053eaa0ac8fa3b40d7e05a33cc8
        with:
          repository: github/docs-early-access
          token: ${{ secrets.DOCUBOT_REPO_PAT }}
          path: docs-early-access

      - name: Merge docs-early-access repo's folders
        run: .github/actions-scripts/merge-early-access.sh

      - uses: ./.github/actions/clone-translations
        with:
          token: ${{ secrets.DOCUBOT_REPO_PAT }}

      - name: 'Build and push image'
        uses: docker/build-push-action@3b5e8027fcad23fda98b2e3ac259d8d67585f671
        with:
          context: .
          push: true
          target: production
          tags: ${{ env.DOCKER_IMAGE }}, ${{ env.DOCKER_IMAGE_CACHE_REF }}
          cache-from: type=registry,ref=${{ env.DOCKER_IMAGE_CACHE_REF }}
          cache-to: type=registry,mode=max,ref=${{ env.DOCKER_IMAGE_CACHE_REF }}
          build-args: |
            BUILD_SHA=${{ github.sha }}

      - name: 'Update docker-compose.prod.yaml template file'
        run: |
          sed 's|#{IMAGE}#|${{ env.DOCKER_IMAGE }}|g' docker-compose.prod.tmpl.yaml > docker-compose.prod.yaml

      - name: 'Apply updated docker-compose.prod.yaml config to canary slot'
        run: |
          az webapp config container set --multicontainer-config-type COMPOSE --multicontainer-config-file docker-compose.prod.yaml --slot canary -n ghdocs-prod -g docs-prod

      # Watch canary slot instances to see when all the instances are ready
      - name: Check that canary slot is ready
        uses: actions/github-script@98814c53be79b1d30f795b907e553d8679345975
        env:
          CHECK_INTERVAL: 10000
        with:
          script: |
            const { execSync } = require('child_process')

            const getStatesForSlot = (slot) => {
              return JSON.parse(
                execSync(
                  `az webapp list-instances --slot ${slot} --query "[].state" -n ghdocs-prod -g docs-prod`,
                  { encoding: 'utf8' }
                )
              )
            }

            let hasStopped = false
            const waitDuration = parseInt(process.env.CHECK_INTERVAL, 10) || 10000
            async function doCheck() {
              const states = getStatesForSlot('canary')
              console.log(`Instance states:`, states)

              // We must wait until at-least 1 instance has STOPPED to know we're looking at the "next" deployment and not the "previous" one
              // That way we don't immediately succeed just because all the previous instances were READY
              if (!hasStopped) {
                hasStopped = states.some((s) => s === 'STOPPED')
              }

              const isAllReady = states.every((s) => s === 'READY')

              if (hasStopped && isAllReady) {
                process.exit(0) // success
              }

              console.log(`checking again in ${waitDuration}ms`)
              setTimeout(doCheck, waitDuration)
            }

            doCheck()

      # TODO - make a request to verify the canary app version aligns with *this* github action workflow commit sha
      - name: 'Swap canary slot to production'
        run: |
          az webapp deployment slot swap --slot canary --target-slot production -n ghdocs-prod -g docs-prod

      - name: Send Slack notification if workflow failed
        uses: someimportantcompany/github-actions-slack-message@1d367080235edfa53df415bd8e0bbab480f29bad
        if: ${{ failure() }}
        with:
          channel: ${{ secrets.DOCS_ALERTS_SLACK_CHANNEL_ID }}
          bot-token: ${{ secrets.SLACK_DOCS_BOT_TOKEN }}
          color: failure
          text: Production deployment (Azure) failed at commit ${{ github.sha }}. See https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}