Merge branch 'master' into nightly-integ-remodel

.github/workflows/docker-nightly-publish.yml

@@ -7,27 +7,18 @@ on:
         description: 'release/nightly/temp, default is nightly'
         required: true
         default: 'nightly'
-      arch:
-        description: 'which images to build [ cpu, cpu-full, pytorch-inf2, pytorch-gpu, tensorrt-llm, lmi, aarch64]'
-        type: string
-        required: false
-        default: '["cpu", "cpu-full", "pytorch-inf2", "pytorch-gpu", "tensorrt-llm", "lmi", "aarch64"]'
   workflow_call:
     inputs:
       mode:
         description: 'release/nightly/temp, default is nightly'
         type: string
         required: true
         default: 'nightly'
-      arch:
-        description: 'which images to build [ cpu, cpu-full, pytorch-inf2, pytorch-gpu, tensorrt-llm, lmi, aarch64]'
-        type: string
-        required: false
-        default: '["cpu", "cpu-full", "pytorch-inf2", "pytorch-gpu", "tensorrt-llm", "lmi", "aarch64"]'
     outputs:
-      djl_version: 
+      djl_version:
         description: "djl version"
         value: ${{ jobs.nightly-build.outputs.djl_version }}
+
 permissions:
   id-token: write
   contents: read
@@ -37,95 +28,65 @@ env:
   DOCKER_HUB_REPO: "deepjavalibrary/djl-serving"
 
 jobs:
-  nightly-build:
-    runs-on: ubuntu-latest
-    strategy:
-      matrix:
-        arch: ${{ fromJson(inputs.arch) }}
-        exclude:
-          - arch: aarch64
-    outputs:
-      djl_version: ${{ steps.get-versions.outputs.DJL_VERSION }}
+  create-runners:
+    runs-on: [ self-hosted, scheduler ]
     steps:
-      - name: Clean disk space
+      - name: Create new CPU instance
+        id: create_cpu_1
         run: |
-          sudo rm -rf \
-          /usr/share/dotnet /usr/local/lib/android /opt/ghc \
-          /usr/local/share/powershell /usr/share/swift /usr/local/.ghcup \
-          $AGENT_TOOLSDIRECTORY
-      - uses: actions/checkout@v4
-      - name: install awscli
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          token=$( 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_cpu $token djl-serving
+      - name: Create new CPU instance
+        id: create_cpu_2
         run: |
-          sudo apt-get update
-          sudo apt-get install awscli -y
-      - 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: Set up JDK 17
-        uses: actions/setup-java@v4
-        with:
-          distribution: 'corretto'
-          java-version: 17
-      - uses: actions/cache@v4
-        with:
-          path: ~/.gradle/caches
-          key: ${{ runner.os }}-gradle-${{ hashFiles('**/*.gradle*') }}
-      - name: Extract DJL and DJL Serving versions from TOML
-        id: get-versions
+          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_cpu $token djl-serving
+      - name: Create new CPU instance
+        id: create_cpu_3
         run: |
-          DJL_VERSION=$(awk -F '=' '/djl / {gsub(/ ?"/, "", $2); print $2}' gradle/libs.versions.toml)
-          SERVING_VERSION=$(awk -F '=' '/serving / {gsub(/ ?"/, "", $2); print $2}' gradle/libs.versions.toml)
-          echo "DJL_VERSION=$DJL_VERSION" >> $GITHUB_ENV
-          echo "SERVING_VERSION=$SERVING_VERSION" >> $GITHUB_ENV
-          echo "DJL_VERSION=$DJL_VERSION" >> $GITHUB_OUTPUT
-      - name: Build release candidate docker image
-        if: ${{ inputs.mode == 'release'}}
-        working-directory: serving/docker
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          token=$( 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_cpu $token djl-serving
+      - name: Create new CPU instance
+        id: create_cpu_4
         run: |
-          export BASE_RELEASE_VERSION="${{ env.SERVING_VERSION }}"
-          export RELEASE_VERSION="${{ env.SERVING_VERSION }}-"
-          docker compose build --no-cache \
-          --build-arg djl_version=${{ env.DJL_VERSION }} \
-          --build-arg djl_serving_version=${{ env.SERVING_VERSION }} \
-          ${{ matrix.arch }}
-      - name: Build temp docker image
-        if: ${{ inputs.mode == '' || inputs.mode == 'temp' || inputs.mode == 'nightly' }}
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          token=$( 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_cpu $token djl-serving
+      - name: Create new CPU instance
+        id: create_cpu_5
         run: |
-          ./gradlew --refresh-dependencies :serving:dockerDeb -Psnapshot
-          cd serving/docker
-          export NIGHTLY="-nightly"
-          echo "NIGHTLY=$NIGHTLY" >> $GITHUB_ENV
-          docker compose build --no-cache \
-          --build-arg djl_version=${{ env.DJL_VERSION }}-SNAPSHOT \
-          --build-arg djl_serving_version=${{ env.SERVING_VERSION }}-SNAPSHOT \
-          ${{ matrix.arch }}
-      - name: Tag and push temp image to ECR repo
-        working-directory: serving/docker
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          token=$( 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_cpu $token djl-serving
+      - name: Create new CPU instance
+        id: create_cpu_6
         run: |
-          ECR_REGION=$(echo "${{ env.AWS_ECR_REPO }}" | awk -F. '{print $4}')
-          aws ecr get-login-password --region $ECR_REGION | docker login --username AWS --password-stdin ${{env.AWS_ECR_REPO}}
-          mode=${{ inputs.mode }}
-          if [ "${{ inputs.mode }}" == "release" ]; then
-            mode=${{ env.DJL_VERSION }}
-          fi
-          tempRunIdTag="${{ env.AWS_ECR_REPO }}:${{ matrix.arch }}-$mode-${GITHUB_RUN_ID}"
-          tempCommitTag="${{ env.AWS_ECR_REPO }}:${{ matrix.arch }}-$mode-${GITHUB_SHA}"
-
-          docker tag ${{ env.DOCKER_HUB_REPO }}:${{ matrix.arch }}${{ env.NIGHTLY }} $tempRunIdTag
-          docker tag ${{ env.DOCKER_HUB_REPO }}:${{ matrix.arch }}${{ env.NIGHTLY }} $tempCommitTag
-          if ${{ inputs.mode == 'nightly' }}; then
-            docker tag ${{ env.DOCKER_HUB_REPO }}:${{ matrix.arch }}${{ env.NIGHTLY }} ${{ env.AWS_ECR_REPO }}:${{ matrix.arch }}-nightly
-          fi
-          time docker push --all-tags  ${{ env.AWS_ECR_REPO }}
-
-  create-runner:
-    if: contains(fromJSON(inputs.arch), 'aarch64')
-    runs-on: [ self-hosted, scheduler ]
-    steps:
-      - name: Create new Graviton instance
-        id: create_aarch64
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          token=$( 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_cpu $token djl-serving
+      - name: Create Graviton instance
+        id: create_graviton_1
         run: |
           cd /home/ubuntu/djl_benchmark_script/scripts
           token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \
@@ -134,24 +95,50 @@ jobs:
           | jq '.token' | tr -d '"' )
           ./start_instance.sh action_graviton $token djl-serving
     outputs:
-      aarch64_instance_id: ${{ steps.create_aarch64.outputs.action_graviton_instance_id }}
+      cpu_instance_id_1: ${{ steps.create_cpu_1.outputs.action_cpu_instance_id }}
+      cpu_instance_id_2: ${{ steps.create_cpu_2.outputs.action_cpu_instance_id }}
+      cpu_instance_id_3: ${{ steps.create_cpu_3.outputs.action_cpu_instance_id }}
+      cpu_instance_id_4: ${{ steps.create_cpu_4.outputs.action_cpu_instance_id }}
+      cpu_instance_id_5: ${{ steps.create_cpu_5.outputs.action_cpu_instance_id }}
+      cpu_instance_id_6: ${{ steps.create_cpu_6.outputs.action_cpu_instance_id }}
+      graviton_instance_id_1: ${{ steps.create_graviton_1.outputs.action_graviton_instance_id }}
 
-  nightly-aarch64:
-    if: contains(fromJSON(inputs.arch), 'aarch64')
+  nightly-build:
+    needs: create-runners
+    strategy:
+      fail-fast: false
+      matrix:
+        containers:
+          - name: cpu
+            instance: cpu
+          - name: cpu-full
+            instance: cpu
+          - name: pytorch-inf2
+            instance: cpu
+          - name: pytorch-gpu
+            instance: cpu
+          - name: tensorrt-llm
+            instance: cpu
+          - name: lmi
+            instance: cpu
+          - name: aarch64
+            instance: aarch64
     runs-on:
       - self-hosted
-      - aarch64
+      - ${{ matrix.containers.instance }}
       - RUN_ID-${{ github.run_id }}
       - RUN_NUMBER-${{ github.run_number }}
       - SHA-${{ github.sha }}
-    timeout-minutes: 60
-    needs: create-runner
+    outputs:
+      djl_version: ${{ steps.get-versions.outputs.DJL_VERSION }}
     steps:
-      - uses: actions/checkout@v4
-      - name: Clean docker env
-        working-directory: serving/docker
+      - name: Clean disk space
         run: |
-          yes | docker system prune -a --volumes
+          sudo rm -rf \
+          /usr/share/dotnet /usr/local/lib/android /opt/ghc \
+          /usr/local/share/powershell /usr/share/swift /usr/local/.ghcup \
+          $AGENT_TOOLSDIRECTORY
+      - uses: actions/checkout@v4
       - name: install awscli
         run: |
           sudo apt-get update
@@ -177,6 +164,7 @@ jobs:
           SERVING_VERSION=$(awk -F '=' '/serving / {gsub(/ ?"/, "", $2); print $2}' gradle/libs.versions.toml)
           echo "DJL_VERSION=$DJL_VERSION" >> $GITHUB_ENV
           echo "SERVING_VERSION=$SERVING_VERSION" >> $GITHUB_ENV
+          echo "DJL_VERSION=$DJL_VERSION" >> $GITHUB_OUTPUT
       - name: Build release candidate docker image
         if: ${{ inputs.mode == 'release' }}
         working-directory: serving/docker
@@ -186,9 +174,9 @@ jobs:
           docker compose build --no-cache \
           --build-arg djl_version=${{ env.DJL_VERSION }} \
           --build-arg djl_serving_version=${{ env.SERVING_VERSION }} \
-          aarch64
+          ${{  matrix.containers.name }}
       - name: Build temp docker image
-        if: ${{  inputs.mode == '' || inputs.mode == 'temp' || inputs.mode == 'nightly' }}
+        if: ${{ inputs.mode == '' || inputs.mode == 'temp' || inputs.mode == 'nightly' }}
         run: |
           ./gradlew --refresh-dependencies :serving:dockerDeb -Psnapshot
           cd serving/docker
@@ -197,37 +185,45 @@ jobs:
           docker compose build --no-cache \
           --build-arg djl_version=${{ env.DJL_VERSION }}-SNAPSHOT \
           --build-arg djl_serving_version=${{ env.SERVING_VERSION }}-SNAPSHOT \
-          aarch64
-      - name: Build and push temp image
+          ${{  matrix.containers.name }}
+      - name: Tag and push temp image to ECR repo
         working-directory: serving/docker
         run: |
-          docker compose build --no-cache \
-          --build-arg djl_version=${{ env.DJL_VERSION }}-SNAPSHOT \
-          --build-arg djl_serving_version=${{ env.SERVING_VERSION }}-SNAPSHOT \
-          aarch64
           ECR_REGION=$(echo "${{ env.AWS_ECR_REPO }}" | awk -F. '{print $4}')
           aws ecr get-login-password --region $ECR_REGION | docker login --username AWS --password-stdin ${{env.AWS_ECR_REPO}}
           mode=${{ inputs.mode }}
           if [ "${{ inputs.mode }}" == "release" ]; then
             mode=${{ env.DJL_VERSION }}
           fi
-          tempRunIdTag="${{ env.AWS_ECR_REPO }}aarch64-$mode-${GITHUB_RUN_ID}"
-          tempCommitTag="${{ env.AWS_ECR_REPO }}:aarch64-$mode-${GITHUB_SHA}"
+          tempRunIdTag="${{ env.AWS_ECR_REPO }}:${{ matrix.containers.name }}-$mode-${GITHUB_RUN_ID}"
+          tempCommitTag="${{ env.AWS_ECR_REPO }}:${{  matrix.containers.name }}-$mode-${GITHUB_SHA}"
 
-          docker tag ${{ env.DOCKER_HUB_REPO }}:aarch64${{ env.NIGHTLY }} $tempRunIdTag
-          docker tag ${{ env.DOCKER_HUB_REPO }}:aarch64${{ env.NIGHTLY }} $tempCommitTag
+          docker tag ${{ env.DOCKER_HUB_REPO }}:${{ matrix.containers.name }}${{ env.NIGHTLY }} $tempRunIdTag
+          docker tag ${{ env.DOCKER_HUB_REPO }}:${{  matrix.containers.name }}${{ env.NIGHTLY }} $tempCommitTag
           if ${{ inputs.mode == 'nightly' }}; then
-            docker tag ${{ env.DOCKER_HUB_REPO }}:aarch64${{ env.NIGHTLY }} ${{ env.AWS_ECR_REPO }}:aarch64-nightly
+            docker tag ${{ env.DOCKER_HUB_REPO }}:${{  matrix.containers.name }}${{ env.NIGHTLY }} ${{ env.AWS_ECR_REPO }}:${{  matrix.containers.name }}-nightly
           fi
           time docker push --all-tags  ${{ env.AWS_ECR_REPO }}
 
-  stop-runner:
+  stop-runners:
     if: always()
     runs-on: [ self-hosted, scheduler ]
-    needs: [nightly-aarch64, create-runner]
+    needs: [nightly-build, create-runners]
     steps:
       - name: Stop all instances
         run: |
           cd /home/ubuntu/djl_benchmark_script/scripts
-          instance_id=${{ needs.create-runner.outputs.aarch64_instance_id }}
+          instance_id=${{ needs.create-runners.outputs.cpu_instance_id_1 }}
+          ./stop_instance.sh $instance_id
+          instance_id=${{ needs.create-runners.outputs.cpu_instance_id_2 }}
+          ./stop_instance.sh $instance_id
+          instance_id=${{ needs.create-runners.outputs.cpu_instance_id_3 }}
+          ./stop_instance.sh $instance_id
+          instance_id=${{ needs.create-runners.outputs.cpu_instance_id_4 }}
+          ./stop_instance.sh $instance_id
+          instance_id=${{ needs.create-runners.outputs.cpu_instance_id_5 }}
+          ./stop_instance.sh $instance_id
+          instance_id=${{ needs.create-runners.outputs.cpu_instance_id_6 }}
+          ./stop_instance.sh $instance_id
+          instance_id=${{ needs.create-runners.outputs.graviton_instance_id_1 }}
           ./stop_instance.sh $instance_id