# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

name: PR

on:
  push:
    branches:
      - "pull-request/[0-9]+"
      # Note: release/* branches are handled by release.yml workflow
  workflow_dispatch:
    inputs:
      run_deploy_operator:
        description: 'Run deploy operator and deployment tests'
        required: false
        type: boolean
        default: false

concurrency:
  # The group name is the ref_name, so that workflows on the same PR/branch have the same group name for cancelling.
  group: docker-build-test-${{ github.ref_name }}
  cancel-in-progress: true

env:
  BUILDER_NAME: b-${{ github.run_id }}-${{ github.run_attempt }}

jobs:

  # ============================================================================
  # SETUP & DETECTION JOBS
  # ============================================================================


  changed-files:
    runs-on: ubuntu-latest
    environment: ${{ github.event_name == 'workflow_dispatch' && 'protected-deploy' || '' }}
    outputs:
      core: ${{ steps.changes.outputs.core }}
      operator: ${{ steps.changes.outputs.operator }}
      deploy: ${{ steps.changes.outputs.deploy }}
      vllm: ${{ steps.changes.outputs.vllm }}
      sglang: ${{ steps.changes.outputs.sglang }}
      trtllm: ${{ steps.changes.outputs.trtllm }}
      builder_name: ${{ steps.export-builder-name.outputs.builder_name }}
    steps:
      - name: Checkout code
        uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955  # v4.3.0
        with:
          fetch-depth: 0
      - name: Check for changes
        id: changes
        uses: ./.github/actions/changed-files
        with:
          gh_token: ${{ github.token }}
      - name: Export builder name
        id: export-builder-name
        run: |
          echo "builder_name=${{ env.BUILDER_NAME }}" >> $GITHUB_OUTPUT

  backend-status-check:
    runs-on: ubuntu-latest
    needs: [changed-files, vllm-pipeline, sglang-pipeline, trtllm-pipeline, operator]  # THIS list determines blocking jobs
    if: always()
    steps:
      - name: "Check all dependent jobs"
        run: |
          echo '${{ toJson(needs) }}' | jq -e 'to_entries | map(.value.result) | all(. as $result | ["success", "skipped"] | any($result == .))'

  deploy-status-check:
    runs-on: ubuntu-latest
    needs: [deploy-operator, deploy-test-vllm, deploy-test-sglang, deploy-test-trtllm]
    if: always()
    steps:
      - name: "Check all deploy test jobs"
        run: |
          echo '${{ toJson(needs) }}' | jq -e 'to_entries | map(.value.result) | all(. as $result | ["success", "skipped"] | any($result == .))'


  # ============================================================================
  # Operator
  # ============================================================================

  operator:
    needs: changed-files
    if: needs.changed-files.outputs.operator == 'true'
    name: Operator
    runs-on: prod-default-v2
    env:
      IMAGE_REGISTRY: ai-dynamo
      IMAGE_REPOSITORY: dynamo
      ECR_HOSTNAME: ${{ secrets.AWS_ACCOUNT_ID }}.dkr.ecr.${{ secrets.AWS_DEFAULT_REGION }}.amazonaws.com
    outputs:
      operator_default_tag: ${{ steps.build-and-push-image.outputs.operator_default_tag }}
    steps:
      - name: Checkout code
        uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955  # v4.3.0
      - name: Initialize Dynamo Builder
        uses: ./.github/actions/init-dynamo-builder
        with:
          builder_name: ${{ needs.changed-files.outputs.builder_name }}
          flavor: general
          all_arch: 'true'
      - name: Docker Login
        uses: ./.github/actions/docker-login
        with:
          aws_default_region: ${{ secrets.AWS_DEFAULT_REGION }}
          aws_account_id: ${{ secrets.AWS_ACCOUNT_ID }}
          azure_acr_hostname: ${{ secrets.AZURE_ACR_HOSTNAME }}
          azure_acr_user: ${{ secrets.AZURE_ACR_USER }}
          azure_acr_password: ${{ secrets.AZURE_ACR_PASSWORD }}
      - name: Linter
        shell: bash
        working-directory: ./deploy/operator
        run: |
          docker buildx build --platform linux/arm64 --target linter --progress=plain --build-arg DOCKER_PROXY=${ECR_HOSTNAME}/dockerhub/ .
      - name: Tester
        shell: bash
        working-directory: ./deploy/operator
        run: |
          docker buildx build --platform linux/arm64 --target tester --progress=plain --build-arg DOCKER_PROXY=${ECR_HOSTNAME}/dockerhub/ .
      - name: Set up Go
        uses: actions/setup-go@44694675825211faa026b3c33043df3e48a5fa00 # v6.0.0
        with:
          go-version: '1.25'
      - name: Set up Python
        uses: actions/setup-python@v5
        with:
          python-version: "3.11"
      - name: Install Python dependencies for operator codegen
        shell: bash
        working-directory: ./deploy/operator
        run: |
          python -m pip install --upgrade pip
          python -m pip install "pydantic>=2,<3" "black==23.1.0" "pyyaml>=6.0"
      - name: Check for uncommitted changes
        shell: bash
        working-directory: ./deploy/operator
        run: |
          make check
      - name: Build and push Container
        id: build-and-push-image
        shell: bash
        working-directory: ./deploy/operator
        env:
          NO_CACHE_FLAG: '' # placeholder for future logic to add no cache flag if needed
        run: |
          ECR_DEFAULT_IMAGE_BASE="${ECR_HOSTNAME}/${IMAGE_REGISTRY}/${IMAGE_REPOSITORY}"
          DEFAULT_TAG="${{ github.sha }}-operator"
          ACR_IMAGE_BASE="${{ secrets.AZURE_ACR_HOSTNAME }}/${IMAGE_REGISTRY}/${IMAGE_REPOSITORY}"
          IMAGE_URIS=(
            "${ECR_DEFAULT_IMAGE_BASE}:${DEFAULT_TAG}"
            "${ACR_IMAGE_BASE}:${DEFAULT_TAG}"
          )

          echo "operator_default_tag=${DEFAULT_TAG}" >> $GITHUB_OUTPUT
          TAGGING_FLAGS=$(printf -- '-t %s ' "${IMAGE_URIS[@]}")
          echo "flags for docker buildx: ${TAGGING_FLAGS}"

          if [[ "$NO_CACHE_FLAG" == "true" ]]; then
            NO_CACHE_FLAG="--no-cache"
          fi
          docker buildx build --push ${NO_CACHE_FLAG} \
              --platform linux/amd64,linux/arm64 \
              --build-arg DOCKER_PROXY=${ECR_HOSTNAME}/dockerhub/ \
              ${TAGGING_FLAGS} -f Dockerfile .

          echo "### 🐳 Operator Container Images" >> $GITHUB_STEP_SUMMARY
          echo "" >> $GITHUB_STEP_SUMMARY
          echo "| Image URI |" >> $GITHUB_STEP_SUMMARY
          echo "|-----|" >> $GITHUB_STEP_SUMMARY
          for image_uri in "${IMAGE_URIS[@]}"; do
            echo "| \`${image_uri}\` |" >> $GITHUB_STEP_SUMMARY
          done


# ============================================================================
# FRAMEWORK PIPELINES (Build → Test → Copy)
# ============================================================================
  # ============================================================================
  # VLLM PIPELINE
  # ============================================================================
  vllm-pipeline:
    name: vllm
    needs: [changed-files]
    if: needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.vllm == 'true' || needs.changed-files.outputs.deploy == 'true'
    uses: ./.github/workflows/build-test-distribute-flavor-matrix.yml
    with:
      framework: vllm
      target: runtime
      platforms: '["amd64", "arm64"]'
      cuda_versions: '["12.9", "13.0"]'
      builder_name: ${{ needs.changed-files.outputs.builder_name }}
      build_timeout_minutes: 60
      copy_timeout_minutes: 10
      run_cpu_only_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.vllm == 'true' }}
      cpu_only_test_markers: 'pre_merge and vllm and gpu_0'
      run_single_gpu_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.vllm == 'true' }}
      single_gpu_test_markers: 'pre_merge and vllm and gpu_1'
      single_gpu_test_timeout_minutes: 35
      run_multi_gpu_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.vllm == 'true' }}
      multi_gpu_test_markers: 'pre_merge and vllm and (gpu_2 or gpu_4)'
      multi_gpu_test_timeout_minutes: 60
    secrets: inherit

  # ============================================================================
  # SGLANG PIPELINE
  # ============================================================================
  sglang-pipeline:
    name: sglang
    needs: [changed-files]
    if: needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.sglang == 'true' || needs.changed-files.outputs.deploy == 'true'
    uses: ./.github/workflows/build-test-distribute-flavor-matrix.yml
    with:
      framework: sglang
      target: runtime
      platforms: '["amd64", "arm64"]'
      cuda_versions: '["12.9", "13.0"]'
      builder_name: ${{ needs.changed-files.outputs.builder_name }}
      build_timeout_minutes: 60
      copy_timeout_minutes: 10
      run_cpu_only_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.sglang == 'true' }}
      cpu_only_test_markers: 'pre_merge and sglang and gpu_0'
      run_single_gpu_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.sglang == 'true' }}
      single_gpu_test_markers: 'pre_merge and sglang and gpu_1'
      run_multi_gpu_tests: false  # all sglang multi-GPU tests are currently skipped; re-enable when fixed
    secrets: inherit

  # ============================================================================
  # TRTLLM PIPELINE
  # ============================================================================
  trtllm-pipeline:
    name: trtllm
    needs: [changed-files]
    if: needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.trtllm == 'true' || needs.changed-files.outputs.deploy == 'true'
    uses: ./.github/workflows/build-test-distribute-flavor-matrix.yml
    with:
      framework: trtllm
      target: runtime
      platforms: '["amd64", "arm64"]'
      cuda_versions: '["13.1"]'
      builder_name: ${{ needs.changed-files.outputs.builder_name }}
      build_timeout_minutes: 60
      copy_timeout_minutes: 10
      run_cpu_only_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.trtllm == 'true' }}
      cpu_only_test_markers: 'pre_merge and trtllm and gpu_0'
      run_single_gpu_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.trtllm == 'true' }}
      single_gpu_test_markers: 'pre_merge and trtllm and gpu_1'
      run_multi_gpu_tests: ${{ needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.trtllm == 'true' }}
      multi_gpu_test_markers: 'pre_merge and trtllm and (gpu_2 or gpu_4)'
      multi_gpu_test_timeout_minutes: 60
    secrets: inherit

  # ============================================================================
  # DEV PIPELINES
  # ============================================================================
  vllm-dev-pipeline:
    name: vllm-dev
    needs: [changed-files]
    if: needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.vllm == 'true'
    uses: ./.github/workflows/build-test-distribute-flavor-matrix.yml
    with:
      framework: vllm
      target: dev
      platforms: '["amd64", "arm64"]'
      cuda_versions: '["12.9", "13.0"]'
      builder_name: ${{ needs.changed-files.outputs.builder_name }}
      build_timeout_minutes: 60
      push_image: false # Only push dev images on main
      copy_to_acr: false
      run_cpu_only_tests: false
      run_single_gpu_tests: false
      run_multi_gpu_tests: false
    secrets: inherit

  sglang-dev-pipeline:
    name: sglang-dev
    needs: [changed-files]
    if: needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.sglang == 'true'
    uses: ./.github/workflows/build-test-distribute-flavor-matrix.yml
    with:
      framework: sglang
      target: dev
      platforms: '["amd64", "arm64"]'
      cuda_versions: '["12.9", "13.0"]'
      builder_name: ${{ needs.changed-files.outputs.builder_name }}
      build_timeout_minutes: 60
      push_image: false # Only push dev images on main
      copy_to_acr: false
      run_cpu_only_tests: false
      run_single_gpu_tests: false
      run_multi_gpu_tests: false
    secrets: inherit

  trtllm-dev-pipeline:
    name: trtllm-dev
    needs: [changed-files]
    if: needs.changed-files.outputs.core == 'true' || needs.changed-files.outputs.trtllm == 'true'
    uses: ./.github/workflows/build-test-distribute-flavor-matrix.yml
    with:
      framework: trtllm
      target: dev
      platforms: '["amd64", "arm64"]'
      cuda_versions: '["13.1"]'
      builder_name: ${{ needs.changed-files.outputs.builder_name }}
      build_timeout_minutes: 60
      push_image: false # Only push dev images on main
      copy_to_acr: false
      run_cpu_only_tests: false
      run_single_gpu_tests: false
      run_multi_gpu_tests: false
    secrets: inherit

   # ============================================================================
   # DEPLOYMENT JOBS
   # Deploy operator and run end-to-end tests on Kubernetes cluster
   # ============================================================================

  deploy-operator:
    runs-on: prod-default-small-v2
    # Run when any deploy test will run: if any framework or deploy files changed
    if: |
      !cancelled() &&
      (needs.changed-files.outputs.vllm == 'true' ||
       needs.changed-files.outputs.sglang == 'true' ||
       needs.changed-files.outputs.trtllm == 'true' ||
       needs.changed-files.outputs.deploy == 'true') &&
      (needs.operator.result == 'success' || needs.operator.result == 'skipped')
    needs: [changed-files, operator]
    outputs:
      NAMESPACE: ${{ steps.namespace.outputs.namespace }}
      OPERATOR_TAG: ${{ steps.operator-tag.outputs.tag }}
    steps:
    - uses: actions/checkout@v4
    - name: Determine operator tag
      id: operator-tag
      run: |
        if [ "${{ needs.operator.result }}" == "success" ]; then
          TAG="${{ needs.operator.outputs.operator_default_tag }}"
        else
          TAG="main-operator"
        fi
        echo "tag=${TAG}" >> $GITHUB_OUTPUT
        echo "Using operator tag: ${TAG}"
    - name: Generate namespace name
      id: namespace
      env:
        BRANCH: ${{ github.ref_name }}
      run: |
        # Sanitize branch name for k8s namespace
        # Invalid patterns: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/
        BRANCH_SANITIZED="${BRANCH//\//-}"
        BRANCH_SANITIZED="${BRANCH_SANITIZED/pull-request/pr}"
        BRANCH_SANITIZED="${BRANCH_SANITIZED//./-}"
        BRANCH_SANITIZED="${BRANCH_SANITIZED:0:10}"
        NAMESPACE="gh-id-${{ github.run_id }}-${BRANCH_SANITIZED}-dt"
        echo "namespace=${NAMESPACE}" >> "$GITHUB_OUTPUT"
    - name: Setup namespace and operator
      uses: ./.github/actions/setup-deploy-namespace
      with:
        kubeconfig_base64: ${{ secrets.AZURE_AKS_CI_KUBECONFIG_B64 }}
        namespace: ${{ steps.namespace.outputs.namespace }}
        registry: ${{ secrets.AZURE_ACR_HOSTNAME }}
        operator_tag: ${{ steps.operator-tag.outputs.tag }}
        hf_token: ${{ secrets.HF_TOKEN }}
   # ============================================================================
   #
   # End-to-end tests for each framework with various deployment profiles
   # ============================================================================
  deploy-test-vllm:
    # !cancelled() && !failure() is required because reusable workflows with skipped
    # internal jobs (e.g. multi-gpu tests) propagate non-success through `needs`,
    # auto-skipping dependents. See: https://github.com/orgs/community/discussions/189172
    if: |
      !cancelled() && !failure() &&
      (needs.changed-files.outputs.vllm == 'true' || needs.changed-files.outputs.deploy == 'true') &&
      (needs.vllm-pipeline.result == 'success')
    runs-on: prod-default-small-v2
    needs: [changed-files, deploy-operator, vllm-pipeline]
    timeout-minutes: 25
    permissions:
      contents: read
    strategy:
      fail-fast: false
      max-parallel: 2
      matrix:
        profile:
          - agg
          - agg_router
          - disagg
          - disagg_router
    name: deploy-test-vllm (${{ matrix.profile }})
    env:
      FRAMEWORK: vllm
    steps:
      - name: Checkout code
        uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955  # v4.3.0
      - name: Run Dynamo Deploy Test
        id: deploy-test
        uses: ./.github/actions/dynamo-deploy-test
        with:
          kubeconfig_base64: ${{ secrets.AZURE_AKS_CI_KUBECONFIG_B64 }}
          namespace: ${{ needs.deploy-operator.outputs.NAMESPACE }}
          registry: ${{ secrets.AZURE_ACR_HOSTNAME }}
          operator_tag: ${{ needs.deploy-operator.outputs.OPERATOR_TAG }}
          hf_token: ${{ secrets.HF_TOKEN }}
          framework: ${{ env.FRAMEWORK }}
          profile: ${{ matrix.profile }}
          image: ${{ secrets.AZURE_ACR_HOSTNAME }}/ai-dynamo/dynamo:${{ github.sha }}-vllm-runtime-cuda12-amd64
          platform_arch: amd64

  deploy-test-sglang:
    runs-on: prod-default-small-v2
    if: |
      !cancelled() && !failure() &&
      (needs.changed-files.outputs.sglang == 'true' || needs.changed-files.outputs.deploy == 'true') &&
      (needs.sglang-pipeline.result == 'success')
    needs: [changed-files, deploy-operator, sglang-pipeline]
    timeout-minutes: 25
    permissions:
      contents: read
    strategy:
      fail-fast: false
      max-parallel: 2
      matrix:
        profile:
          - agg
          - agg_router
    name: deploy-test-sglang (${{ matrix.profile }})
    env:
      FRAMEWORK: sglang
    steps:
      - name: Checkout code
        uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955  # v4.3.0
      - name: Run Dynamo Deploy Test
        id: deploy-test
        uses: ./.github/actions/dynamo-deploy-test
        with:
          kubeconfig_base64: ${{ secrets.AZURE_AKS_CI_KUBECONFIG_B64 }}
          namespace: ${{ needs.deploy-operator.outputs.NAMESPACE }}
          registry: ${{ secrets.AZURE_ACR_HOSTNAME }}
          operator_tag: ${{ needs.deploy-operator.outputs.OPERATOR_TAG }}
          hf_token: ${{ secrets.HF_TOKEN }}
          framework: ${{ env.FRAMEWORK }}
          profile: ${{ matrix.profile }}
          image: ${{ secrets.AZURE_ACR_HOSTNAME }}/ai-dynamo/dynamo:${{ github.sha }}-sglang-runtime-cuda12-amd64
          platform_arch: amd64

  deploy-test-trtllm:
    runs-on: prod-default-small-v2
    if: |
      !cancelled() && !failure() &&
      (needs.changed-files.outputs.trtllm == 'true' || needs.changed-files.outputs.deploy == 'true') &&
      (needs.trtllm-pipeline.result == 'success')
    needs: [changed-files, deploy-operator, trtllm-pipeline]
    timeout-minutes: 25
    permissions:
      contents: read
    strategy:
      fail-fast: false
      max-parallel: 2
      matrix:
        profile:
          - agg
          - agg_router
          # Disabled: trtllm disagg profiles consistently timeout (~32 min) with 0% success rate.
          # Re-enable once the underlying disagg deployment issue is resolved.
          # - disagg
          # - disagg_router
    name: deploy-test-trtllm (${{ matrix.profile }})
    env:
      FRAMEWORK: trtllm
    steps:
      - name: Checkout code
        uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955  # v4.3.0
      - name: Run Dynamo Deploy Test
        id: deploy-test
        uses: ./.github/actions/dynamo-deploy-test
        with:
          kubeconfig_base64: ${{ secrets.AZURE_AKS_CI_KUBECONFIG_B64 }}
          namespace: ${{ needs.deploy-operator.outputs.NAMESPACE }}
          registry: ${{ secrets.AZURE_ACR_HOSTNAME }}
          operator_tag: ${{ needs.deploy-operator.outputs.OPERATOR_TAG }}
          hf_token: ${{ secrets.HF_TOKEN }}
          framework: ${{ env.FRAMEWORK }}
          profile: ${{ matrix.profile }}
          image: ${{ secrets.AZURE_ACR_HOSTNAME }}/ai-dynamo/dynamo:${{ github.sha }}-trtllm-runtime-cuda13-amd64
          platform_arch: amd64

  # ============================================================================
  # CLEANUP JOBS
  # Clean up ephemeral Kubernetes namespace and resources
  # ============================================================================

  clean-k8s-builder:
    name: Clean K8s builder if exists
    runs-on: prod-default-small-v2
    if: always()
    needs: [vllm-pipeline, sglang-pipeline, trtllm-pipeline, operator, changed-files]
    steps:
    - name: Checkout repository
      uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955  # v4.3.0
    - name: Create K8s builders (skip bootstrap)
      uses: ./.github/actions/bootstrap-buildkit
      continue-on-error: true
      with:
        builder_name: ${{ needs.changed-files.outputs.builder_name }}
        buildkit_worker_addresses: '' # k8s builder
        skip_bootstrap: true
    - name: Builder Cleanup in case of k8s builder
      shell: bash
      run: |
        docker buildx rm ${{ needs.changed-files.outputs.builder_name }} || true


  cleanup:
    name: Cleanup AKS resources
    runs-on: prod-default-small-v2
    if: always()
    needs: [deploy-operator, deploy-test-trtllm, deploy-test-sglang, deploy-test-vllm]
    steps:
    - name: Checkout code
      uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955  # v4.3.0
    - name: Teardown namespace
      if: needs.deploy-operator.outputs.NAMESPACE != ''
      uses: ./.github/actions/teardown-deploy-namespace
      with:
        kubeconfig_base64: ${{ secrets.AZURE_AKS_CI_KUBECONFIG_B64 }}
        namespace: ${{ needs.deploy-operator.outputs.NAMESPACE }}