DocsGPT/.github/workflows/docker-image-verify.yml

name: Verify the Docker image works offline

# Builds the backend image and runs its offline check with networking off, so
# a change that reintroduces a first-request download (a tokenizer, tiktoken's
# encoding, an embedding model) fails here instead of in an air-gapped install.
# Then starts the standalone Compose stack on the image and checks that the one
# published port serves both the API and the web UI.

on:
  workflow_dispatch:
  pull_request:
    paths:
      - 'docsgpt/Dockerfile'
      - '.dockerignore'
      - 'application/**'
      - 'docsgpt/requirements*.txt'
      - 'docsgpt/scripts/prefetch_models.py'
      - 'docsgpt/scripts/verify_offline.py'
      - 'docsgpt/vectorstore/model_registry.py'
      - 'docsgpt/parser/tokenization.py'
      - 'docsgpt/vectorstore/embeddings_local.py'
      - 'docsgpt/ui.py'
      - 'frontend/**'
      - 'scripts/build_frontend.sh'
      - 'deployment/docker-compose-standalone.yaml'
      - '.github/workflows/docker-image-verify.yml'

permissions:
  contents: read

jobs:
  verify:
    strategy:
      matrix:
        # "" is the slim default; "-docling" bakes docling, its models and
        # tesseract in, so the conversion check in verify_offline runs too.
        variant: ["", "-docling"]
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
        with:
          persist-credentials: false

      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f # v3.12.0

      - name: Build the image
        uses: docker/build-push-action@10e90e3645eae34f1e60eeb005ba3a3d33f178e8 # v6.19.2
        with:
          file: ./docsgpt/Dockerfile
          context: .
          platforms: linux/amd64
          load: true
          # The name the standalone Compose file runs, under a tag no registry has.
          tags: arc53/docsgpt:verify${{ matrix.variant }}
          build-args: |
            EXTRAS=${{ matrix.variant == '-docling' && 'docling' || '' }}
            INSTALL_TESSERACT=${{ matrix.variant == '-docling' && 'true' || 'false' }}
          cache-from: type=gha,scope=verify${{ matrix.variant }}
          cache-to: type=gha,mode=max,scope=verify${{ matrix.variant }}

      - name: Image size
        env:
          IMAGE: arc53/docsgpt:verify${{ matrix.variant }}
        run: |
          docker image inspect "$IMAGE" --format '{{.Size}}' | awk '{printf "uncompressed: %.2f GB\n", $1/1e9}'
          docker history "$IMAGE" --format '{{.Size}}\t{{.CreatedBy}}' | head -20

      - name: Offline verification (no network)
        env:
          IMAGE: arc53/docsgpt:verify${{ matrix.variant }}
        run: |
          docker run --rm --network none "$IMAGE" \
            python -m docsgpt.scripts.verify_offline

      - name: The standalone stack serves the API and the UI on one port
        env:
          DOCSGPT_IMAGE_TAG: verify
          DOCSGPT_IMAGE_VARIANT: ${{ matrix.variant }}
        run: |
          set -euo pipefail
          # --pull missing keeps the image built above; postgres and redis are pulled.
          docker compose -f deployment/docker-compose-standalone.yaml up -d --pull missing backend
          base=http://127.0.0.1:7091
          for _ in $(seq 1 90); do
            if curl -fsS "$base/api/health" >/dev/null 2>&1; then break; fi
            sleep 2
          done
          curl -fsS "$base/api/health"
          echo
          curl -fsS "$base/" | grep -q 'src="/config.js"'
          curl -fsS "$base/config.js" | grep -q 'window.__DOCSGPT_ENV__'
          # A client-side route falls back to the UI's index.html.
          curl -fsS "$base/settings" | grep -q 'src="/config.js"'
          echo "API and UI served on $base"

      - name: Stack logs
        if: failure()
        env:
          DOCSGPT_IMAGE_TAG: verify
          DOCSGPT_IMAGE_VARIANT: ${{ matrix.variant }}
        run: docker compose -f deployment/docker-compose-standalone.yaml logs --no-color

      - name: Stop the stack
        if: always()
        env:
          DOCSGPT_IMAGE_TAG: verify
          DOCSGPT_IMAGE_VARIANT: ${{ matrix.variant }}
        run: docker compose -f deployment/docker-compose-standalone.yaml down -v