build-production-site.yaml

name: Build production site

on:
  release:
    types: [published]

# This workflow builds the site with Hugo, compresses it, uploads it to an
# object storage bucket, and updates the sandbox Algolia app to reflect the
# content in the production version of the site. Another non-github.com delivery
# system is responsible for deploying the compressed site to a production
# webserver.
#
# This workflow is triggered whenever a release tag matching the `vX.Y.Z`
# format is created on the docs repo.
#
# The working directory for this GitHub Action is under:
# /home/runner/work/docs/docs/
#
# For this workflow, two repos are checked out:
# - The github.com/linode/docs repo is checked out into the `docs-repo` folder
# - The github.com/linode/linode-docs-theme repo is checked out into the
#   `linode-docs-theme` folder.
#
# This workflow runs `hugo` to build the site inside the `docs-repo` folder.
# The site is rendered into the `docs-repo/public/` directory. The workflow
# uses SFTP to upload a tarball of this folder over to the destination web
# servers.
#
# We check out the linode-docs-theme repo because it contains a
# `linode_algolia_admin` tool that we use to update the Algolia search indices.
#
# Inside the docs repo, the go.mod file contains a reference to which git
# commit of the linode-docs-theme is associated with the docs repo.
# We inspect this file to get that commit hash before checking out the
# linode-docs-theme repo.
#
# The following is a diagram of the above-mentioned repos and files. It does not
# show all the files within those repositories:
#
# .
#
# ├── docs-repo
# │   ├── go.mod
# │   └── public
# └── linode-docs-theme-repo
#     └── scripts
#         └── linode_algolia_admin

jobs:
  build-production-site:
    if: github.repository_owner == 'linode'
    runs-on: ubuntu-latest

    environment:
      name: build_production_site
      url: ${{ vars.DOCS_WEBSITE_URL }}

    steps:
    - name: 1. Set up Hugo
      uses: peaceiris/actions-hugo@v2
      with:
        hugo-version: ${{ vars.HUGO_VERSION }}

    - name: 2. Checkout docs repo
      uses: actions/checkout@v3
      with:
        path: 'docs-repo'
        # ref: 'main'

      # Debugging step so we can check if the right version
      # of the docs repo is checked out
    - name: (Debugging info) Print current docs repo branch/ref/commit
      working-directory: ./docs-repo
      run: |
        git status
        git log -1

    - name: 3. Set up SSH agent (linode-docs-theme repo deploy key)
      uses: webfactory/ssh-agent@v0.7.0
      with:
        ssh-private-key: ${{ secrets.LINODE_DOCS_THEME_DEPLOY_KEY_FOR_DOCS_DEPLOY_GHA }}

    - name: 4. Clone docs theme repo
      run: |
        git clone git@github.com:$GITHUB_REPOSITORY_OWNER/linode-docs-theme linode-docs-theme-repo

    - name: 5. Get linode-docs-theme commit hash from docs go.mod
      id: get-theme-version
      working-directory: ./docs-repo
      run: |
        # `hugo mod graph` prints output that looks like:
        # project github.com/linode/linode-docs-theme@v0.0.0-20230706164309-cc2dbdeb7143+vendor
        # In this example, cc2dbdeb7143 is the relevant commit hash for the
        # linode-docs-theme repo. The following commands grab the commit hash
        # from the `hugo mod graph` string

        LINODE_DOCS_THEME_VERSION=$(hugo mod graph | grep linode-docs-theme | cut -d '+' -f 1 | grep -o '[^-]*$')
        echo "VERSION=$LINODE_DOCS_THEME_VERSION" >> $GITHUB_OUTPUT

    - name: 6. Check out linode-docs-theme commit hash from docs go.mod
      working-directory: ./linode-docs-theme-repo
      run: |
        git checkout ${{ steps.get-theme-version.outputs.VERSION }}

      # Debugging step so we can check if the right version
      # of the Algolia admin tool is checked out
    - name: (Debugging info) Print Linode Algolia admin tool version
      working-directory: ./linode-docs-theme-repo
      run: |
        cd scripts/linode_algolia_admin/
        go run main.go version

      # Debugging step that lists the Hugo-generated images
      # directory *before* the imache cache restore step
    - name: (Debugging info) List contents of images dir
      continue-on-error: true
      run: ls -al ${{ vars.HUGO_IMAGE_CACHE_PATH }}

    - name: 7. Restore Hugo generated images cache
      uses: ylemkimon/cache-restore@v2
      with:
        path: ${{ vars.HUGO_IMAGE_CACHE_PATH }}
        key: ${{ vars.HUGO_IMAGE_CACHE_NAME }}
        restore-keys: ${{ vars.HUGO_IMAGE_CACHE_NAME }}

      # Debugging step that lists the Hugo-generated images
      # directory *after* the imache cache restore step,
      # to make sure that the restore happened successfully
    - name: (Debugging info) List contents of images dir
      continue-on-error: true
      run: ls -al ${{ vars.HUGO_IMAGE_CACHE_PATH }}

    - name: 8. Install Node.js
      uses: actions/setup-node@v3
      with:
        node-version: ${{ vars.NODE_VERSION }}

    - name: 9. Install dependencies (Node)
      working-directory: ./docs-repo
      run: npm ci

    - name: 10. Build Hugo
      env:
        HUGOxPARAMSxSEARCH_CONFIG2xAPP_ID: ${{ vars.ALGOLIA_APP_ID }}
        HUGOxPARAMSxSEARCH_CONFIG2xAPI_KEY: ${{ vars.ALGOLIA_SEARCH_KEY }}
      working-directory: ./docs-repo
      run: |
        hugo config
        hugo -b "${{ vars.DOCS_WEBSITE_URL }}" --gc --minify -d public

      # Debugging step that lists the Hugo-rendered files
      # to make sure the site was built
    - name: (Debugging info) List rendered files
      working-directory: ./docs-repo
      run: |
        sudo apt install -y tree
        tree -L 1 public

      # When Hugo builds, it renders search index data to a collection of
      # data files under public/. This step uploads that data to the
      # Algolia search backend
    - name: 11. Update Algolia
      working-directory: ./linode-docs-theme-repo
      env:
        ALGOLIA_APP_ID: ${{ vars.ALGOLIA_APP_ID }}
        ALGOLIA_ADMIN_API_KEY: ${{ secrets.ALGOLIA_WRITE_KEY }}
      run: |
        cd scripts/linode_algolia_admin/

        # Hugo renders the site to the `public` directory within the docs repo.
        # This includes three files with Algolia index data:
        # - public/index.json
        # - public/api/index.json
        # - public/data/sections/index.json
        # The pushdocs command uploads this data to Algolia
        go run main.go pushdocs -source-dir ../../../docs-repo/public

        # The settings (e.g. ranking configuration) and synonyms for the
        # linode-merged index are stored in version control, and the
        # following command pushes those settings/synonyms to the Algolia app
        go run main.go -indices linode-merged push -replace settings,synonyms

        # The refresh-merge-and-push sequence assumes that the linode-wp and linode-community indices already exist in the Algolia app
        go run main.go sequence refresh-merge-and-push

      # Why build Hugo twice? In order to render the Explore Docs nav menu,
      # Hugo downloads data from the Algolia search backend. However, the
      # first time we build Hugo in this workflow, that data isn't up-to-date
      # with new content we may be publishing. So, we build Hugo a second time
      # after we have performed the previous Update Algolia step.
      # In summary:
      # - Build Hugo the first time in order to render the JSON search index
      #   data files
      # - Update Algolia with those data files
      # - Build Hugo again so that the navigation UI can be rendered with that
      #   updated info from Algolia
      # It's a little redundant, but solves the chicken-or-egg problem
    - name: 12. Build Hugo (second time)
      env:
        HUGOxPARAMSxSEARCH_CONFIG2xAPP_ID: ${{ vars.ALGOLIA_APP_ID }}
        HUGOxPARAMSxSEARCH_CONFIG2xAPI_KEY: ${{ vars.ALGOLIA_SEARCH_KEY }}
      working-directory: ./docs-repo
      run: |
        hugo config
        hugo -b "${{ vars.DOCS_WEBSITE_URL }}" --gc --minify -d public

      # The gitcommithash.txt file is used when the site is deployed to the
      # webserver to verify that the deployment was successful.
    - name: 13. Add gitcommithash.txt to rendered public/ folder
      working-directory: ./docs-repo
      run: |
        echo $GITHUB_SHA > public/gitcommithash.txt

      # Make a tarball of the site, because it will upload much, much quicker
      # than the uncompressed rendered site. The commit for this workflow run
      # is encoded in the name of the tarball.
    - name: 14. Create tarball of docs website
      run: |
        tar --owner=${{ secrets.DOCS_WEBSITE_USER }} --group=${{ secrets.DOCS_WEBSITE_USER }} -czf docs-$GITHUB_SHA.tar.gz -C docs-repo/public/ .

    - name: 15. Set up S3cmd CLI tool
      uses: s3-actions/s3cmd@v1.5.0
      with:
        provider: linode
        region: ${{ secrets.OBJ_REGION }}
        access_key: ${{ secrets.OBJ_ACCESS_KEY }}
        secret_key: ${{ secrets.OBJ_SECRET_KEY }}

      # Upload the tarball to an object storage bucket. Another non-github.com
      # delivery system is used to deploy this tarball to the prod webserver
    - name: 16. Upload website tarball to object storage
      run: |
        echo "$(date +%s)" > build-date.txt # Print current Unix timestamp to a text file

        s3cmd put docs-$GITHUB_SHA.tar.gz s3://${{ secrets.OBJ_BUCKET_NAME }}/prod/docs-$GITHUB_SHA/
        s3cmd put build-date.txt s3://${{ secrets.OBJ_BUCKET_NAME }}/prod/docs-$GITHUB_SHA/