lmcache--lmcache/.github/workflows/cpu_device.yml

name: CPU Device Tests

# CPU device tests: server bench + vLLM e2e. Both job types share a
# single environment (built once per OS) to avoid re-installing vLLM.

on:
  workflow_call:
  workflow_dispatch:
  pull_request:
    branches:
      - "dev"
      - "release-**"

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

permissions:
  contents: read

defaults:
  run:
    shell: bash

jobs:
  changes:
    name: Detect changes
    runs-on: ubuntu-latest
    permissions:
      pull-requests: read
    outputs:
      lmcache: >-
        ${{
          github.event_name == 'workflow_dispatch' ||
          github.event_name == 'workflow_call' ||
          steps.filter.outputs.lmcache == 'true'
        }}
    steps:
      - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
        if: github.event_name == 'push'
        with:
          fetch-depth: 2

      - uses: dorny/paths-filter@v3
        if: >-
          github.event_name == 'push' ||
          github.event_name == 'pull_request'
        id: filter
        with:
          filters: |
            lmcache:
              - 'lmcache/**'
              - 'pyproject.toml'
              - 'setup.py'
              - 'requirements/**.txt'
              - '.github/workflows/cpu_device.yml'
              - '.github/scripts/cpu_device_test.sh'
              - '.github/scripts/cpu_server_bench_test.sh'
              - '.github/scripts/cpu_vllm_e2e_test.sh'
              - '.github/scripts/install_vllm_cpu.sh'
              - '.github/scripts/install_lmcache_cpu.sh'
              - '.github/scripts/download_model.sh'
              - '.github/scripts/run-cpu-e2e-validation.sh'
              - '!operator/**'

  cpu-device-test:
    needs: changes
    if: needs.changes.outputs.lmcache == 'true'
    name: CPU device (${{ matrix.os }})
    runs-on: ${{ matrix.os }}
    timeout-minutes: 120
    strategy:
      fail-fast: false
      matrix:
        os: [ubuntu-22.04, macos-latest]

    steps:
      - name: Harden Runner
        uses: step-security/harden-runner@20cf305ff2072d973412fa9b1e3a4f227bda3c76 # v2.14.0
        with:
          egress-policy: audit

      - name: Checkout
        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
        with:
          fetch-depth: 1

      - name: Setup Python 3.12
        uses: actions/setup-python@83679a892e2d95755f2dac6acb0bfd1e9ac5d548 # v6.1.0
        with:
          python-version: "3.12"
          cache: pip
          cache-dependency-path: |
            pyproject.toml
            requirements/*.txt

      - name: Install FFmpeg (macOS)
        if: runner.os == 'macOS'
        run: |
          if ! command -v ffmpeg >/dev/null 2>&1; then
            brew install ffmpeg
          else
            echo "ffmpeg already installed: $(ffmpeg -version | head -n1)"
          fi

      - name: Cache opt-125m model (GitHub release)
        uses: actions/cache@v4
        with:
          path: ~/.cache/huggingface
          key: opt-125m-gh-v2

      - name: Install vLLM CPU (prebuilt nightly from PyPI)
        run: bash .github/scripts/install_vllm_cpu.sh

      - name: Install lmcache (CPU-only, no vLLM)
        run: bash .github/scripts/install_lmcache_cpu.sh

      - name: Download facebook/opt-125m (GitHub release)
        run: bash .github/scripts/download_opt125m_github.sh

      - name: Server bench — LMCache-driven
        run: |
          LMCACHE_BENCH_TRANSFER_MODE=lmcache_driven \
          LMCACHE_HTTP_PORT_BENCH=18080 \
          LMCACHE_ZMQ_PORT_BENCH=15555 \
            bash .github/scripts/cpu_device_test.sh server_bench

      - name: Server bench — Engine-driven
        run: |
          LMCACHE_BENCH_TRANSFER_MODE=engine_driven \
          LMCACHE_HTTP_PORT_BENCH=18081 \
          LMCACHE_ZMQ_PORT_BENCH=15556 \
            bash .github/scripts/cpu_device_test.sh server_bench

      - name: vLLM e2e — Engine-driven (pickle)
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          LMCACHE_E2E_TRANSPORT_MODE=pickle \
          LMCACHE_HTTP_PORT_E2E=18082 \
          LMCACHE_ZMQ_PORT_E2E=15557 \
          VLLM_PORT_E2E=18000 \
            bash .github/scripts/cpu_device_test.sh vllm_e2e

      # POSIX SHM transport is Linux-only.
      - name: vLLM e2e — Engine-driven (shm)
        if: runner.os == 'Linux'
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          LMCACHE_E2E_TRANSPORT_MODE=shm \
          LMCACHE_HTTP_PORT_E2E=18083 \
          LMCACHE_ZMQ_PORT_E2E=15558 \
          VLLM_PORT_E2E=18001 \
            bash .github/scripts/cpu_device_test.sh vllm_e2e

      - name: vLLM e2e — Engine-driven
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          LMCACHE_E2E_TRANSPORT_MODE=engine_driven \
          LMCACHE_HTTP_PORT_E2E=18084 \
          LMCACHE_ZMQ_PORT_E2E=15559 \
          VLLM_PORT_E2E=18002 \
            bash .github/scripts/cpu_device_test.sh vllm_e2e

      - name: Upload logs on failure
        if: failure()
        uses: actions/upload-artifact@v4
        with:
          name: cpu-device-logs-${{ matrix.os }}
          path: |
            /tmp/cpu_device_*.log
          if-no-files-found: ignore