Skip to content

AWS Long Run Tests on GPU #1

AWS Long Run Tests on GPU

AWS Long Run Tests on GPU #1

name: AWS Long Run Tests on GPU
env:
CARGO_TERM_COLOR: always
ACTION_RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
RUSTFLAGS: "-C target-cpu=native"
RUST_BACKTRACE: "full"
RUST_MIN_STACK: "8388608"
SLACK_CHANNEL: ${{ secrets.SLACK_CHANNEL }}
SLACK_ICON: https://pbs.twimg.com/profile_images/1274014582265298945/OjBKP9kn_400x400.png
SLACK_USERNAME: ${{ secrets.BOT_USERNAME }}
SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
on:
# Allows you to run this workflow manually from the Actions tab as an alternative.
workflow_dispatch:
schedule:
# Weekly tests will be triggered each Friday at 1a.m.
- cron: '0 1 * * FRI'
jobs:
setup-instance:
name: Setup instance (gpu-tests)
if: github.event_name != 'schedule' ||
(github.event_name == 'schedule' && github.repository == 'zama-ai/tfhe-rs')
runs-on: ubuntu-latest
outputs:
runner-name: ${{ steps.start-instance.outputs.label }}
steps:
- name: Start instance
id: start-instance
uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
with:
mode: start
github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
slab-url: ${{ secrets.SLAB_BASE_URL }}
job-secret: ${{ secrets.JOB_SECRET }}
backend: hyperstack
profile: single-h100
cuda-tests:
name: Long run GPU H100 tests
needs: [ setup-instance ]
concurrency:
group: ${{ github.workflow }}_${{github.event_name}}_${{ github.ref }}
cancel-in-progress: true
runs-on: ${{ needs.setup-instance.outputs.runner-name }}
strategy:
fail-fast: false
# explicit include-based build matrix, of known valid options
matrix:
include:
- os: ubuntu-22.04
cuda: "12.2"
gcc: 11
env:
CUDA_PATH: /usr/local/cuda-${{ matrix.cuda }}
CMAKE_VERSION: 3.29.6
steps:
# Mandatory on hyperstack since a bootable volume is not re-usable yet.
- name: Install dependencies
run: |
sudo apt update
sudo apt install -y checkinstall zlib1g-dev libssl-dev libclang-dev
wget https://github.com/Kitware/CMake/releases/download/v${{ env.CMAKE_VERSION }}/cmake-${{ env.CMAKE_VERSION }}.tar.gz
tar -zxvf cmake-${{ env.CMAKE_VERSION }}.tar.gz
cd cmake-${{ env.CMAKE_VERSION }}
./bootstrap
make -j"$(nproc)"
sudo make install
- name: Checkout tfhe-rs
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
- name: Set up home
run: |
echo "HOME=/home/ubuntu" >> "${GITHUB_ENV}"
- name: Install latest stable
uses: dtolnay/rust-toolchain@7b1c307e0dcbda6122208f10795a713336a9b35a
with:
toolchain: stable
- name: Export CUDA variables
if: ${{ !cancelled() }}
run: |
echo "CUDA_PATH=$CUDA_PATH" >> "${GITHUB_ENV}"
echo "$CUDA_PATH/bin" >> "${GITHUB_PATH}"
echo "LD_LIBRARY_PATH=$CUDA_PATH/lib:$LD_LIBRARY_PATH" >> "${GITHUB_ENV}"
echo "CUDACXX=/usr/local/cuda-${{ matrix.cuda }}/bin/nvcc" >> "${GITHUB_ENV}"
# Specify the correct host compilers
- name: Export gcc and g++ variables
if: ${{ !cancelled() }}
run: |
{
echo "CC=/usr/bin/gcc-${{ matrix.gcc }}";
echo "CXX=/usr/bin/g++-${{ matrix.gcc }}";
echo "CUDAHOSTCXX=/usr/bin/g++-${{ matrix.gcc }}";
echo "HOME=/home/ubuntu";
} >> "${GITHUB_ENV}"
- name: Check device is detected
if: ${{ !cancelled() }}
run: nvidia-smi
- name: Run tests
run: |
make test_integer_long_run_gpu
slack-notify:
name: Slack Notification
needs: [ setup-instance, cuda-tests ]
runs-on: ubuntu-latest
if: ${{ always() && needs.cuda-tests.result != 'skipped' && failure() }}
continue-on-error: true
steps:
- name: Send message
uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
env:
SLACK_COLOR: ${{ needs.cuda-tests.result }}
SLACK_MESSAGE: "Integer GPU H100 long run tests finished with status: ${{ needs.cuda-tests.result }}. (${{ env.ACTION_RUN_URL }})"
teardown-instance:
name: Teardown instance (gpu-tests)
if: ${{ always() && needs.setup-instance.result != 'skipped' }}
needs: [ setup-instance, cuda-tests ]
runs-on: ubuntu-latest
steps:
- name: Stop instance
id: stop-instance
uses: zama-ai/slab-github-runner@801df0b8db5ea2b06128b7476c652f5ed5f193a8
with:
mode: stop
github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
slab-url: ${{ secrets.SLAB_BASE_URL }}
job-secret: ${{ secrets.JOB_SECRET }}
label: ${{ needs.setup-instance.outputs.runner-name }}
- name: Slack Notification
if: ${{ failure() }}
continue-on-error: true
uses: rtCamp/action-slack-notify@c33737706dea87cd7784c687dadc9adf1be59990
env:
SLACK_COLOR: ${{ job.status }}
SLACK_MESSAGE: "Instance teardown (gpu-long-run-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"