name: Nightly Test (Intel)

on:
  schedule:
    # 22:15 IST (UTC+5:30) → 16:45 UTC daily.
    - cron: '45 16 * * *'
  push:
    branches:
      - main
    paths:
      - "python/sglang/version.py"
  workflow_dispatch:
    inputs:
      continue_on_error:
        description: 'Continue on error (do not fail the workflow on test failures)'
        required: false
        type: boolean
        default: true
  workflow_call:
    inputs:
      ref:
        description: "Branch, tag or SHA to checkout"
        required: false
        type: string
        default: ""
      continue_on_error:
        description: 'Continue on error (do not fail the workflow on test failures)'
        required: false
        type: boolean
        default: true

concurrency:
  group: nightly-test-intel-${{ github.event_name == 'workflow_dispatch' && format('manual-{0}', github.run_id) || inputs.ref && format('caller-{0}', github.run_id) || github.ref }}
  cancel-in-progress: ${{ !inputs.ref && github.event_name != 'workflow_call' && github.event_name != 'workflow_dispatch' }}

jobs:
  nightly-xpu-1-gpu:
    if: github.repository == 'sgl-project/sglang'
    runs-on: intel-bmg-nightly
    env:
      DOCKERHUB_INTEL_USERNAME: ${{ secrets.DOCKERHUB_INTEL_USERNAME }}
      DOCKERHUB_INTEL_TOKEN: ${{ secrets.DOCKERHUB_INTEL_TOKEN }}
    steps:
      - name: Reset workspace ownership
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true

      - name: Checkout code
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          ref: ${{ inputs.ref || github.sha }}

      - name: Start CI container (pull intel/sglang-dev:latest)
        run: |
          export HF_TOKEN="$(cat ~/huggingface_token.txt)"
          bash scripts/ci/xpu/xpu_ci_start_container.sh
        env:
          GITHUB_WORKSPACE: ${{ github.workspace }}

      - name: Overlay in-tree sglang test helpers onto image site-packages
        run: |
          # Bridges the up-to-24h window between merge and nightly image rebuild.
          dest=$(docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
                 "import sglang, os; print(os.path.dirname(sglang.__file__))")
          docker exec ci_sglang_xpu cp \
            /sglang-checkout/python/sglang/test/ci/ci_utils.py \
            "$dest/test/ci/ci_utils.py"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
            "import sglang.test.ci.ci_utils; print('overlay ok')"

      - name: HF login + install run_suite extras
        timeout-minutes: 10
        run: |
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir tabulate
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir --no-deps xgrammar==0.1.33
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "lm-eval==0.4.9"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir pytest
          # Diffusion extras (needed by test_xpu_flux2_dev, test_xpu_zimage_turbo).
          # Pins mirror python/pyproject_xpu.toml [project.optional-dependencies.diffusion].
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "sglang[diffusion]"
          # HF /whoami-v2 is aggressively rate-limited and the three matrix
          # jobs (1/2/4-GPU) all hit it at the same second under a scheduled
          # run. Retry with linear backoff + jitter to desynchronize them.
          docker exec ci_sglang_xpu /bin/bash -c '
            for i in 1 2 3 4 5; do
              /opt/venv/bin/hf auth login --token "${HF_TOKEN}" && exit 0
              sleep_secs=$((i * 15 + RANDOM % 15))
              echo "hf auth login failed (attempt $i), retrying in ${sleep_secs}s..."
              sleep "${sleep_secs}"
            done
            exit 1
          '

      - name: Download olmOCR-bench dataset (for test_deepseek_ocr_2_olmbench)
        timeout-minutes: 30
        run: |
          # PyMuPDF renders the bench PDFs to images; bench_sglang.py errors
          # every sample without it.
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir pymupdf
          docker exec ci_sglang_xpu /bin/bash -c '
            /opt/venv/bin/hf download --repo-type dataset allenai/olmOCR-bench \
              --local-dir /sglang-checkout/olmOCR-bench'

      - name: Nightly Test (1-GPU XPU)
        timeout-minutes: 240
        run: |
          touch github_summary.md nightly-xpu-1-gpu-metrics.jsonl
          docker exec ci_sglang_xpu bash -c "
            source /opt/venv/bin/activate &&
            cd /sglang-checkout/test &&
            OLMOCR_BENCH_DIR=/sglang-checkout/olmOCR-bench/bench_data \
            GITHUB_STEP_SUMMARY=/sglang-checkout/github_summary.md \
            SGLANG_IS_IN_CI=true \
            SGLANG_TEST_METRICS_FILE=/sglang-checkout/nightly-xpu-1-gpu-metrics.jsonl \
            python3 run_suite.py --hw xpu --suite nightly-xpu-1-gpu --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
          " || TEST_EXIT_CODE=$?
          echo "$(<github_summary.md)" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

      - name: Upload per-model metrics jsonl
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: nightly-xpu-1-gpu-metrics
          path: nightly-xpu-1-gpu-metrics.jsonl
          if-no-files-found: warn
          retention-days: 30

      - name: Cleanup container
        if: always()
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true
          rm -rf test/result.jsonl test/results test/.pytest_cache .pytest_cache || true
          find . -type d -name "__pycache__" -prune -exec rm -rf {} + || true
          find . -type f -name "*.pyc" -delete || true
          docker rm -f ci_sglang_xpu || true
          if [[ -n "${CI_SGLANG_XPU_IMAGE:-}" ]]; then
            docker rmi -f "${CI_SGLANG_XPU_IMAGE}" || true
          fi

  nightly-xpu-2-gpu:
    if: github.repository == 'sgl-project/sglang'
    runs-on: intel-bmg-nightly
    env:
      DOCKERHUB_INTEL_USERNAME: ${{ secrets.DOCKERHUB_INTEL_USERNAME }}
      DOCKERHUB_INTEL_TOKEN: ${{ secrets.DOCKERHUB_INTEL_TOKEN }}
    steps:
      - name: Reset workspace ownership
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true

      - name: Checkout code
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          ref: ${{ inputs.ref || github.sha }}

      - name: Start CI container (pull intel/sglang-dev:latest)
        run: |
          export HF_TOKEN="$(cat ~/huggingface_token.txt)"
          bash scripts/ci/xpu/xpu_ci_start_container.sh
        env:
          GITHUB_WORKSPACE: ${{ github.workspace }}

      - name: Overlay in-tree sglang test helpers onto image site-packages
        run: |
          # Bridges the up-to-24h window between merge and nightly image rebuild.
          dest=$(docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
                 "import sglang, os; print(os.path.dirname(sglang.__file__))")
          docker exec ci_sglang_xpu cp \
            /sglang-checkout/python/sglang/test/ci/ci_utils.py \
            "$dest/test/ci/ci_utils.py"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
            "import sglang.test.ci.ci_utils; print('overlay ok')"

      - name: HF login + install run_suite extras
        timeout-minutes: 10
        run: |
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir tabulate
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir --no-deps xgrammar==0.1.33
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "lm-eval==0.4.9"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir pytest
          # Diffusion extras (needed by test_xpu_flux2_dev, test_xpu_zimage_turbo).
          # Pins mirror python/pyproject_xpu.toml [project.optional-dependencies.diffusion].
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "sglang[diffusion]"
          # HF /whoami-v2 is aggressively rate-limited and the three matrix
          # jobs (1/2/4-GPU) all hit it at the same second under a scheduled
          # run. Retry with linear backoff + jitter to desynchronize them.
          docker exec ci_sglang_xpu /bin/bash -c '
            for i in 1 2 3 4 5; do
              /opt/venv/bin/hf auth login --token "${HF_TOKEN}" && exit 0
              sleep_secs=$((i * 15 + RANDOM % 15))
              echo "hf auth login failed (attempt $i), retrying in ${sleep_secs}s..."
              sleep "${sleep_secs}"
            done
            exit 1
          '

      - name: Nightly Test (2-GPU XPU)
        timeout-minutes: 240
        run: |
          touch github_summary.md nightly-xpu-2-gpu-metrics.jsonl
          docker exec ci_sglang_xpu bash -c "
            source /opt/venv/bin/activate &&
            cd /sglang-checkout/test &&
            GITHUB_STEP_SUMMARY=/sglang-checkout/github_summary.md \
            SGLANG_IS_IN_CI=true \
            SGLANG_TEST_METRICS_FILE=/sglang-checkout/nightly-xpu-2-gpu-metrics.jsonl \
            python3 run_suite.py --hw xpu --suite nightly-xpu-2-gpu --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
          " || TEST_EXIT_CODE=$?
          echo "$(<github_summary.md)" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

      - name: Upload per-model metrics jsonl
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: nightly-xpu-2-gpu-metrics
          path: nightly-xpu-2-gpu-metrics.jsonl
          if-no-files-found: warn
          retention-days: 30

      - name: Cleanup container
        if: always()
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true
          rm -rf test/result.jsonl test/results test/.pytest_cache .pytest_cache || true
          find . -type d -name "__pycache__" -prune -exec rm -rf {} + || true
          find . -type f -name "*.pyc" -delete || true
          docker rm -f ci_sglang_xpu || true
          if [[ -n "${CI_SGLANG_XPU_IMAGE:-}" ]]; then
            docker rmi -f "${CI_SGLANG_XPU_IMAGE}" || true
          fi

  nightly-xpu-4-gpu:
    if: github.repository == 'sgl-project/sglang'
    runs-on: intel-bmg-nightly
    env:
      DOCKERHUB_INTEL_USERNAME: ${{ secrets.DOCKERHUB_INTEL_USERNAME }}
      DOCKERHUB_INTEL_TOKEN: ${{ secrets.DOCKERHUB_INTEL_TOKEN }}
    steps:
      - name: Reset workspace ownership
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true

      - name: Checkout code
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          ref: ${{ inputs.ref || github.sha }}

      - name: Start CI container (pull intel/sglang-dev:latest)
        run: |
          export HF_TOKEN="$(cat ~/huggingface_token.txt)"
          bash scripts/ci/xpu/xpu_ci_start_container.sh
        env:
          GITHUB_WORKSPACE: ${{ github.workspace }}

      - name: Overlay in-tree sglang test helpers onto image site-packages
        run: |
          # Bridges the up-to-24h window between merge and nightly image rebuild.
          dest=$(docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
                 "import sglang, os; print(os.path.dirname(sglang.__file__))")
          docker exec ci_sglang_xpu cp \
            /sglang-checkout/python/sglang/test/ci/ci_utils.py \
            "$dest/test/ci/ci_utils.py"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
            "import sglang.test.ci.ci_utils; print('overlay ok')"

      - name: HF login + install run_suite extras
        timeout-minutes: 10
        run: |
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir tabulate
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir --no-deps xgrammar==0.1.33
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "lm-eval==0.4.9"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir pytest
          # Diffusion extras (needed by test_xpu_flux2_dev, test_xpu_zimage_turbo).
          # Pins mirror python/pyproject_xpu.toml [project.optional-dependencies.diffusion].
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "sglang[diffusion]"
          # HF /whoami-v2 is aggressively rate-limited and the three matrix
          # jobs (1/2/4-GPU) all hit it at the same second under a scheduled
          # run. Retry with linear backoff + jitter to desynchronize them.
          docker exec ci_sglang_xpu /bin/bash -c '
            for i in 1 2 3 4 5; do
              /opt/venv/bin/hf auth login --token "${HF_TOKEN}" && exit 0
              sleep_secs=$((i * 15 + RANDOM % 15))
              echo "hf auth login failed (attempt $i), retrying in ${sleep_secs}s..."
              sleep "${sleep_secs}"
            done
            exit 1
          '

      - name: Nightly Test (4-GPU XPU)
        timeout-minutes: 480
        run: |
          touch github_summary.md nightly-xpu-4-gpu-metrics.jsonl
          docker exec ci_sglang_xpu bash -c "
            source /opt/venv/bin/activate &&
            cd /sglang-checkout/test &&
            GITHUB_STEP_SUMMARY=/sglang-checkout/github_summary.md \
            SGLANG_IS_IN_CI=true \
            SGLANG_TEST_METRICS_FILE=/sglang-checkout/nightly-xpu-4-gpu-metrics.jsonl \
            python3 run_suite.py --hw xpu --suite nightly-xpu-4-gpu --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
          " || TEST_EXIT_CODE=$?
          echo "$(<github_summary.md)" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

      - name: Upload per-model metrics jsonl
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: nightly-xpu-4-gpu-metrics
          path: nightly-xpu-4-gpu-metrics.jsonl
          if-no-files-found: warn
          retention-days: 30

      - name: Cleanup container
        if: always()
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true
          rm -rf test/result.jsonl test/results test/.pytest_cache .pytest_cache || true
          find . -type d -name "__pycache__" -prune -exec rm -rf {} + || true
          find . -type f -name "*.pyc" -delete || true
          docker rm -f ci_sglang_xpu || true
          if [[ -n "${CI_SGLANG_XPU_IMAGE:-}" ]]; then
            docker rmi -f "${CI_SGLANG_XPU_IMAGE}" || true
          fi

  nightly-xpu-8-gpu:
    if: github.repository == 'sgl-project/sglang'
    runs-on: intel-bmg-nightly
    env:
      DOCKERHUB_INTEL_USERNAME: ${{ secrets.DOCKERHUB_INTEL_USERNAME }}
      DOCKERHUB_INTEL_TOKEN: ${{ secrets.DOCKERHUB_INTEL_TOKEN }}
    steps:
      - name: Reset workspace ownership
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true

      - name: Checkout code
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          ref: ${{ inputs.ref || github.sha }}

      - name: Start CI container (pull intel/sglang-dev:latest)
        run: |
          export HF_TOKEN="$(cat ~/huggingface_token.txt)"
          bash scripts/ci/xpu/xpu_ci_start_container.sh
        env:
          GITHUB_WORKSPACE: ${{ github.workspace }}

      - name: Overlay in-tree sglang test helpers onto image site-packages
        run: |
          # Bridges the up-to-24h window between merge and nightly image rebuild.
          dest=$(docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
                 "import sglang, os; print(os.path.dirname(sglang.__file__))")
          docker exec ci_sglang_xpu cp \
            /sglang-checkout/python/sglang/test/ci/ci_utils.py \
            "$dest/test/ci/ci_utils.py"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -c \
            "import sglang.test.ci.ci_utils; print('overlay ok')"

      - name: HF login + install run_suite extras
        timeout-minutes: 10
        run: |
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir tabulate
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir --no-deps xgrammar==0.1.33
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "lm-eval==0.4.9"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir pytest
          # Diffusion extras
          # Pins mirror python/pyproject_xpu.toml [project.optional-dependencies.diffusion].
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir "sglang[diffusion]"
          # HF /whoami-v2 is aggressively rate-limited and the three matrix
          # jobs (1/2/4/8-GPU) all hit it at the same second under a scheduled
          # run. Retry with linear backoff + jitter to desynchronize them.
          docker exec ci_sglang_xpu /bin/bash -c '
            for i in 1 2 3 4 5; do
              /opt/venv/bin/hf auth login --token "${HF_TOKEN}" && exit 0
              sleep_secs=$((i * 15 + RANDOM % 15))
              echo "hf auth login failed (attempt $i), retrying in ${sleep_secs}s..."
              sleep "${sleep_secs}"
            done
            exit 1
          '

      - name: Nightly Test (8-GPU XPU)
        timeout-minutes: 480
        run: |
          touch github_summary.md nightly-xpu-8-gpu-metrics.jsonl
          docker exec ci_sglang_xpu bash -c "
            source /opt/venv/bin/activate &&
            cd /sglang-checkout/test &&
            GITHUB_STEP_SUMMARY=/sglang-checkout/github_summary.md \
            SGLANG_IS_IN_CI=true \
            SGLANG_TEST_METRICS_FILE=/sglang-checkout/nightly-xpu-8-gpu-metrics.jsonl \
            python3 run_suite.py --hw xpu --suite nightly-xpu-8-gpu --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }}
          " || TEST_EXIT_CODE=$?
          echo "$(<github_summary.md)" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

      - name: Upload per-model metrics jsonl
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: nightly-xpu-8-gpu-metrics
          path: nightly-xpu-8-gpu-metrics.jsonl
          if-no-files-found: warn
          retention-days: 30

      - name: Cleanup container
        if: always()
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true
          rm -rf test/result.jsonl test/results test/.pytest_cache .pytest_cache || true
          find . -type d -name "__pycache__" -prune -exec rm -rf {} + || true
          find . -type f -name "*.pyc" -delete || true
          docker rm -f ci_sglang_xpu || true
          if [[ -n "${CI_SGLANG_XPU_IMAGE:-}" ]]; then
            docker rmi -f "${CI_SGLANG_XPU_IMAGE}" || true
          fi

  check-all-jobs:
    if: always() && (github.repository == 'sgl-project/sglang' || github.event_name == 'workflow_dispatch')
    needs:
      - nightly-xpu-1-gpu
      - nightly-xpu-2-gpu
      - nightly-xpu-4-gpu
      - nightly-xpu-8-gpu
    runs-on: ubuntu-latest
    steps:
      - name: Check if any job failed
        run: |
          if [[ "${{ contains(needs.*.result, 'failure') }}" == "true" ]]; then
            echo "One or more nightly test jobs failed"
            exit 1
          fi
          if [[ "${{ contains(needs.*.result, 'cancelled') }}" == "true" ]]; then
            echo "One or more nightly test jobs were cancelled"
            exit 1
          fi
          echo "All nightly test jobs passed"