Integration tests #1374
Workflow file for this run
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Integration tests | |
| on: | |
| workflow_dispatch: | |
| inputs: | |
| djl-version: | |
| description: 'The released version of DJL.' | |
| required: false | |
| default: '' | |
| tag-suffix: | |
| description: 'Run tests on the specific tags suffix i.e. arch-{suffix}' | |
| required: false | |
| type: string | |
| default: 'nightly' | |
| workflow_call: | |
| inputs: | |
| djl-version: | |
| description: 'The released version of DJL.' | |
| required: false | |
| type: string | |
| default: 'nightly' | |
| tag-suffix: | |
| description: 'Run tests on the specific tags suffix i.e. arch-{suffix}' | |
| required: false | |
| type: string | |
| default: '' | |
| outputs: | |
| failure_cpu: | |
| value: ${{ jobs.test.outputs.failure_cpu || '0' }} | |
| failure_gpu: | |
| value: ${{ jobs.test.outputs.failure_gpu || '0' }} | |
| failure_aarch64: | |
| value: ${{ jobs.test.outputs.failure_aarch64 || '0' }} | |
| failure_lmi: | |
| value: ${{ jobs.test.outputs.failure_lmi || '0' }} | |
| failure_trtllm: | |
| value: ${{ jobs.test.outputs.failure_trtllm || '0' }} | |
| failure_neuron: | |
| value: ${{ jobs.test.outputs.failure_neuron || jobs.transformers-neuronx-container-unit-tests.outputs.failure || '0' }} | |
| permissions: | |
| id-token: write | |
| contents: read | |
| env: | |
| AWS_ECR_REPO: "185921645874.dkr.ecr.us-east-1.amazonaws.com/djl-ci-temp" | |
| jobs: | |
| create-runners: | |
| runs-on: [self-hosted, scheduler] | |
| steps: | |
| - name: Create new G6 instance | |
| id: create_gpu | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \ | |
| https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \ | |
| --fail \ | |
| | jq '.token' | tr -d '"' ) | |
| ./start_instance.sh action_g6 $token djl-serving | |
| - name: Create new G6 instance | |
| id: create_gpu2 | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \ | |
| https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \ | |
| --fail \ | |
| | jq '.token' | tr -d '"' ) | |
| ./start_instance.sh action_g6 $token djl-serving | |
| - name: Create new G6 instance | |
| id: create_gpu3 | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \ | |
| https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \ | |
| --fail \ | |
| | jq '.token' | tr -d '"' ) | |
| ./start_instance.sh action_g6 $token djl-serving | |
| - name: Create new G6 instance | |
| id: create_gpu4 | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \ | |
| https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \ | |
| --fail \ | |
| | jq '.token' | tr -d '"' ) | |
| ./start_instance.sh action_g6 $token djl-serving | |
| - name: Create new Graviton instance | |
| id: create_aarch64 | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \ | |
| https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \ | |
| --fail \ | |
| | jq '.token' | tr -d '"' ) | |
| ./start_instance.sh action_graviton $token djl-serving | |
| - name: Create new Inf2.24xl instance | |
| id: create_inf2 | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \ | |
| https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \ | |
| --fail \ | |
| | jq '.token' | tr -d '"' ) | |
| ./start_instance.sh action_inf2 $token djl-serving | |
| - name: Create new Inf2.24xl instance | |
| id: create_inf2_2 | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \ | |
| https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \ | |
| --fail \ | |
| | jq '.token' | tr -d '"' ) | |
| ./start_instance.sh action_inf2 $token djl-serving | |
| outputs: | |
| gpu_instance_id_1: ${{ steps.create_gpu.outputs.action_g6_instance_id }} | |
| gpu_instance_id_2: ${{ steps.create_gpu2.outputs.action_g6_instance_id }} | |
| gpu_instance_id_3: ${{ steps.create_gpu3.outputs.action_g6_instance_id }} | |
| gpu_instance_id_4: ${{ steps.create_gpu4.outputs.action_g6_instance_id }} | |
| aarch64_instance_id: ${{ steps.create_aarch64.outputs.action_graviton_instance_id }} | |
| inf2_instance_id_1: ${{ steps.create_inf2.outputs.action_inf2_instance_id }} | |
| inf2_instance_id_2: ${{ steps.create_inf2_2.outputs.action_inf2_instance_id }} | |
| test: | |
| runs-on: | |
| - ${{ matrix.test.gh-runner && matrix.test.instance || 'self-hosted' }} | |
| - ${{ matrix.test.gh-runner && matrix.test.instance || format('RUN_ID-{0}', github.run_id) }} | |
| - ${{ matrix.test.gh-runner && matrix.test.instance || format('RUN_NUMBER-{0}', github.run_number) }} | |
| - ${{ matrix.test.gh-runner && matrix.test.instance || format('SHA-{0}', github.sha) }} | |
| - ${{ matrix.test.instance }} | |
| timeout-minutes: 90 | |
| needs: create-runners | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| test: | |
| - test: TestCpuFull | |
| instance: ubuntu-latest | |
| gh-runner: true | |
| failure-prefix: cpu | |
| - test: TestCpuBoth | |
| instance: ubuntu-latest | |
| gh-runner: true | |
| failure-prefix: cpu | |
| - test: TestGpu | |
| instance: g6 | |
| failure-prefix: gpu | |
| - test: TestAarch64 | |
| instance: aarch64 | |
| failure-prefix: aarch64 | |
| - test: TestHfHandler | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestTrtLlmHandler1 | |
| instance: g6 | |
| failure-prefix: trtllm | |
| - test: TestTrtLlmHandler2 | |
| instance: g6 | |
| failure-prefix: trtllm | |
| - test: TestSchedulerSingleGPU | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestSchedulerMultiGPU | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestLmiDist1 | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestLmiDist2 | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestVllm1 | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestVllmLora | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestLmiDistLora | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestNeuronx1 | |
| instance: inf2 | |
| failure-prefix: neuron | |
| - test: TestNeuronx2 | |
| instance: inf2 | |
| failure-prefix: neuron | |
| - test: TestNeuronxRollingBatch | |
| instance: inf2 | |
| failure-prefix: neuron | |
| - test: TestMultiModal | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestTextEmbedding | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestLmiDistPipelineParallel | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestLmiDistMultiNode | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestCorrectnessTrtLlm | |
| instance: g6 | |
| failure-prefix: trtllm | |
| - test: TestCorrectnessLmiDist | |
| instance: g6 | |
| failure-prefix: lmi | |
| - test: TestCorrectnessNeuronx | |
| instance: inf2 | |
| failure-prefix: neuron | |
| outputs: | |
| failure_cpu: ${{ steps.test-failure.outputs.failure_cpu }} | |
| failure_gpu: ${{ steps.test-failure.outputs.failure_gpu }} | |
| failure_aarch64: ${{ steps.test-failure.outputs.failure_aarch64 }} | |
| failure_lmi: ${{ steps.test-failure.outputs.failure_lmi }} | |
| failure_trtllm: ${{ steps.test-failure.outputs.failure_trtllm }} | |
| failure_neuron: ${{ steps.test-failure.outputs.failure_neuron }} | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - name: Clean env | |
| run: | | |
| yes | docker system prune -a --volumes | |
| sudo rm -rf /home/ubuntu/actions-runner/_work/_tool/Java_Corretto_jdk/ | |
| echo "wait dpkg lock..." | |
| while sudo fuser /var/{lib/{dpkg,apt/lists},cache/apt/archives}/lock >/dev/null 2>&1; do sleep 5; done | |
| - name: Set up Python3 | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: '3.10.x' | |
| - name: Install pip dependencies | |
| run: | | |
| python -m pip install --upgrade pip | |
| pip install pytest requests "numpy<2" pillow huggingface_hub awscli torch | |
| - name: Install awscurl | |
| working-directory: tests/integration | |
| run: | | |
| wget https://publish.djl.ai/awscurl/awscurl | |
| chmod +x awscurl | |
| mkdir outputs | |
| - name: Configure AWS Credentials | |
| if: matrix.test.instance == 'ubuntu-latest' | |
| uses: aws-actions/configure-aws-credentials@v4 | |
| with: | |
| role-to-assume: arn:aws:iam::185921645874:role/github-actions-djl-serving | |
| aws-region: us-east-1 | |
| - name: Test | |
| working-directory: tests/integration | |
| env: | |
| TEST_DJL_VERSION: ${{ inputs.djl-version }} | |
| OVERRIDE_IMAGE_TAG_SUFFIX: ${{ inputs.tag-suffix }} | |
| IMAGE_REPO: ${{ env.AWS_ECR_REPO }} | |
| run: | | |
| ECR_REGION=$(echo "${{ env.AWS_ECR_REPO }}" | awk -F. '{print $4}') | |
| aws ecr get-login-password --region $ECR_REGION | docker login --username AWS --password-stdin ${{env.AWS_ECR_REPO}} | |
| python -m pytest -s -k ${{ matrix.test.test }} tests.py | |
| - name: Cleanup | |
| working-directory: tests/integration | |
| run: | | |
| rm -rf outputs | |
| rm awscurl | |
| - name: On Failure | |
| id: test-failure | |
| if: ${{ failure() }} | |
| working-directory: tests/integration | |
| run: | | |
| for file in outputs/*; do if [ -f "$file" ]; then echo "Contents of $file:"; cat "$file"; echo; fi; done | |
| sudo rm -rf outputs && sudo rm -rf models | |
| rm awscurl | |
| ./remove_container.sh | |
| failure_prefix="${{ matrix.test.failure-prefix }}" | |
| echo "failure_${failure_prefix}=1" >> "$GITHUB_OUTPUT" | |
| - name: Upload test logs | |
| if: ${{ always() }} | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: test-${{ matrix.test.test }}-logs | |
| path: tests/integration/all_logs/ | |
| transformers-neuronx-container-unit-tests: | |
| runs-on: | |
| - self-hosted | |
| - inf2 | |
| - RUN_ID-${{ github.run_id }} | |
| - RUN_NUMBER-${{ github.run_number }} | |
| - SHA-${{ github.sha }} | |
| timeout-minutes: 15 | |
| needs: create-runners | |
| outputs: | |
| failure: ${{ steps.failure.outputs.failure }} | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - name: Clean env | |
| run: | | |
| yes | docker system prune -a --volumes | |
| sudo rm -rf /home/ubuntu/actions-runner/_work/_tool/Java_Corretto_jdk/ | |
| echo "wait dpkg lock..." | |
| while sudo fuser /var/{lib/{dpkg,apt/lists},cache/apt/archives}/lock >/dev/null 2>&1; do sleep 5; done | |
| - name: Set up Python3 | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: '3.10.x' | |
| - name: Install pip dependencies | |
| run: pip3 install requests numpy pillow wheel | |
| - name: Configure AWS Credentials | |
| uses: aws-actions/configure-aws-credentials@v4 | |
| with: | |
| role-to-assume: arn:aws:iam::185921645874:role/github-actions-djl-serving | |
| aws-region: us-east-1 | |
| - name: Download models and dockers | |
| run: | | |
| if [ "${{ github.event.inputs.djl-version }}" == "temp" ]; then | |
| DOCKER_IMAGE_URI="${{ env.AWS_ECR_REPO }}:pytorch-inf2-temp-${GITHUB_SHA}" | |
| elif [ -n "${{ inputs.tag-suffix }}" ]; then | |
| DOCKER_IMAGE_URI="${{ env.AWS_ECR_REPO }}:pytorch-inf2-${{ inputs.tag-suffix }}" | |
| else | |
| DOCKER_IMAGE_URI="${{ env.AWS_ECR_REPO }}:pytorch-inf2-nightly" | |
| fi | |
| echo "DOCKER_IMAGE_URI=$DOCKER_IMAGE_URI" >>$GITHUB_ENV | |
| ECR_REGION=$(echo "${{ env.AWS_ECR_REPO }}" | awk -F. '{print $4}') | |
| aws ecr get-login-password --region $ECR_REGION | docker login --username AWS --password-stdin ${{env.AWS_ECR_REPO}} | |
| echo $DOCKER_IMAGE_URI | |
| docker pull $DOCKER_IMAGE_URI | |
| - name: Run djl_python unit/integration tests on container | |
| working-directory: engines/python/setup | |
| run: | | |
| # Setup | |
| pip install setuptools | |
| python3 -m setup bdist_wheel | |
| mkdir logs | |
| docker run -t --rm --network="host" \ | |
| --name neuron-test \ | |
| -v $PWD/:/opt/ml/model/ \ | |
| -w /opt/ml/model \ | |
| --device=/dev/neuron0:/dev/neuron0 \ | |
| $DOCKER_IMAGE_URI \ | |
| /bin/bash -c "'pip install /opt/ml/model/dist/*.whl pytest' && \ | |
| pytest djl_python/tests/neuron_test_scripts/ | tee logs/results.log" | |
| # Cleanup | |
| sudo rm -rf TinyLlama .pytest_cache djl_python | |
| # Fail on failed tests | |
| if grep -F "failed" logs/results.log &>/dev/null; then exit 1; fi | |
| - name: On fail step | |
| id: failure | |
| if: ${{ failure() }} | |
| working-directory: engines/python/setup | |
| run: | | |
| cat logs/results.log | |
| echo "failure=1" >> "$GITHUB_OUTPUT" | |
| - name: Upload test logs | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: transformers-neuronx-${{ matrix.arch }}-logs | |
| path: engines/python/setup/logs/ | |
| stop-runners: | |
| if: always() | |
| runs-on: [ self-hosted, scheduler ] | |
| needs: [ create-runners, test, transformers-neuronx-container-unit-tests] | |
| steps: | |
| - name: Stop all instances | |
| run: | | |
| cd /home/ubuntu/djl_benchmark_script/scripts | |
| instance_id=${{ needs.create-runners.outputs.gpu_instance_id_1 }} | |
| ./stop_instance.sh $instance_id | |
| instance_id=${{ needs.create-runners.outputs.gpu_instance_id_2 }} | |
| ./stop_instance.sh $instance_id | |
| instance_id=${{ needs.create-runners.outputs.gpu_instance_id_3 }} | |
| ./stop_instance.sh $instance_id | |
| instance_id=${{ needs.create-runners.outputs.gpu_instance_id_4 }} | |
| ./stop_instance.sh $instance_id | |
| instance_id=${{ needs.create-runners.outputs.aarch64_instance_id }} | |
| ./stop_instance.sh $instance_id | |
| instance_id=${{ needs.create-runners.outputs.inf2_instance_id_1 }} | |
| ./stop_instance.sh $instance_id | |
| instance_id=${{ needs.create-runners.outputs.inf2_instance_id_2 }} | |
| ./stop_instance.sh $instance_id |