integration_test.yml

name: Integration tests

on:
  pull_request_review:
    branches: [ main ]
    types: [submitted]

  workflow_dispatch:

jobs:
  model-pack-tests:
    if: github.event.review.state == 'approved' || github.event_name == 'workflow_dispatch'
    runs-on: [ubuntu-latest]
    strategy:
      fail-fast: false
    steps:
      - name: Git checkout
        uses: actions/checkout@v4
        with:
          lfs: true

      - name: Set up Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'
          cache: pip

      - name: create venv and install dependencies
        run: |
          rm -r /tmp/kazu-env || true

          PIP_CACHE_DIR=$(pip cache dir)
          echo "pip cache dir is "${PIP_CACHE_DIR}
          python -m venv /tmp/kazu-env
          . /tmp/kazu-env/bin/activate
          python -m pip install --upgrade pip
          pip install --upgrade --upgrade-strategy eager --index-url https://download.pytorch.org/whl/cpu "torch>=2.0.0"
          pip install -e ."[dev]" --cache-dir $PIP_CACHE_DIR --upgrade --upgrade-strategy eager

      - name: Set up JDK 11
        uses: actions/setup-java@v3
        with:
          java-version: '11'
          distribution: 'temurin'

# TODO: network issue on gradle DL
#      - name: Setup and execute Gradle 'test' task
#        uses: gradle/gradle-build-action@v2
#        with:
#          arguments: test
#          build-root-directory: kazu-jvm/

      - name: Build model pack caches and test
        id: model_pack_cache_build
        run: |
          export BASE_KAZU_CONF_PATH=$GITHUB_WORKSPACE"/kazu/conf"
          # set PYTHONUNBUFFERED so that we see failures in logs before freeze
          export PYTHONUNBUFFERED=1

          # random suffix to prevent clashes with other runs
          export MODEL_PACK_BUILD_PATH=$GITHUB_WORKSPACE"/temp_model_pack_test_dir_"$RANDOM
          export KAZU_LOGGING_CONF=$GITHUB_WORKSPACE"/kazu/utils/model_pack_build_logging.conf"
          export KAZU_MODEL_PACK_BUILD_RESOURCES_PATH=$GITHUB_WORKSPACE"/resources/"

          mkdir $MODEL_PACK_BUILD_PATH

          . /tmp/kazu-env/bin/activate

          python -m kazu.utils.build_and_test_model_packs \
          --base_configuration_path $BASE_KAZU_CONF_PATH \
          --model_packs_to_build kazu_model_pack_public \
          --model_pack_output_path $MODEL_PACK_BUILD_PATH \
          --skip_tests \
          --logging_config_path $KAZU_LOGGING_CONF \
          --debug

          export KAZU_MODEL_PACK=$(echo $MODEL_PACK_BUILD_PATH/kazu_model_pack_public)

          # skip server tests on runner due to low shared memory.
          export SKIP_KAZU_SERVER_TESTS=1
          # skip experimental modules
          export SKIP_KAZU_EXPERIMENTAL_TESTS=1


          # separate out hypothesis and non-hypothesis tests
          # because the hypothesis cache causes memory to grow significantly and
          # cause the run to fail.
          pytest -m "not hypothesis"
          pytest -m "hypothesis"
          # build sphinx html and run doctests
          # this needs to be after the model pack build because they depend on the
          # model pack being present for doctests that load the full pipeline.
          # We've already built the html docs in the pre-commit and docs
          # check, but doctest can only be done here. It seems worth it to double
          # check and build the html target while sphinx is running, just in case
          # it interacts weirdly with the model pack build somehow.
          make -C docs html doctest

          # need to call twice, for NFS bug
          rm -r /tmp/kazu-env || true
          rm -r /tmp/kazu-env || true

          rm -r $MODEL_PACK_BUILD_PATH || true
          rm -r $MODEL_PACK_BUILD_PATH || true