Files
tfhe-rs/.github/workflows/gpu_integer_long_run_tests.yml
David Testé 9918dacd6a chore(ci): change workflow jobs naming convention
The term "bpr" means Branch Protection Rule. It helps one to
identify any job that must pass before being able to merge to the
base branch.
2025-09-10 15:36:45 +02:00

128 lines
4.5 KiB
YAML

name: gpu_integer_long_run_tests
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 }}
CHECKOUT_TOKEN: ${{ secrets.REPO_CHECKOUT_TOKEN || secrets.GITHUB_TOKEN }}
IS_PR: ${{ github.event_name == 'pull_request' }}
on:
# Allows you to run this workflow manually from the Actions tab as an alternative.
workflow_dispatch:
schedule:
# Nightly tests will be triggered each evening 8p.m.
- cron: "0 20 * * *"
pull_request:
permissions:
contents: read
jobs:
setup-instance:
name: gpu_integer_long_run_tests/setup-instance
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@79939325c3c429837c10d6041e4fd8589d328bac
with:
mode: start
github-token: ${{ secrets.SLAB_ACTION_TOKEN }}
slab-url: ${{ secrets.SLAB_BASE_URL }}
job-secret: ${{ secrets.JOB_SECRET }}
backend: hyperstack
profile: multi-gpu-test
cuda-tests:
name: gpu_integer_long_run_tests/cuda-tests
needs: [ setup-instance ]
concurrency:
group: ${{ github.workflow_ref }}_${{github.event_name}}
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: "12.8"
gcc: 11
timeout-minutes: 4320 # 72 hours
steps:
- name: Checkout tfhe-rs
uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8
with:
persist-credentials: 'false'
token: ${{ env.CHECKOUT_TOKEN }}
- name: Setup Hyperstack dependencies
uses: ./.github/actions/gpu_setup
with:
cuda-version: ${{ matrix.cuda }}
gcc-version: ${{ matrix.gcc }}
- name: Install latest stable
uses: dtolnay/rust-toolchain@e97e2d8cc328f1b50210efc529dca0028893a2d9 # zizmor: ignore[stale-action-refs] this action doesn't create releases
with:
toolchain: stable
- name: Enable nvidia multi-process service
run: |
nvidia-cuda-mps-control -d
- name: Run tests
run: |
if [[ "${IS_PR}" == "true" ]]; then
make test_integer_short_run_gpu
else
make test_integer_long_run_gpu
fi
slack-notify:
name: gpu_integer_long_run_tests/slack-notify
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@e31e87e03dd19038e411e38ae27cbad084a90661
env:
SLACK_COLOR: ${{ needs.cuda-tests.result }}
SLACK_MESSAGE: "Integer GPU long run tests finished with status: ${{ needs.cuda-tests.result }}. (${{ env.ACTION_RUN_URL }})"
teardown-instance:
name: gpu_integer_long_run_tests/teardown-instance
if: ${{ always() && needs.setup-instance.result == 'success' }}
needs: [ setup-instance, cuda-tests ]
runs-on: ubuntu-latest
steps:
- name: Stop instance
id: stop-instance
uses: zama-ai/slab-github-runner@79939325c3c429837c10d6041e4fd8589d328bac
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@e31e87e03dd19038e411e38ae27cbad084a90661
env:
SLACK_COLOR: ${{ job.status }}
SLACK_MESSAGE: "Instance teardown (gpu-long-run-tests) finished with status: ${{ job.status }}. (${{ env.ACTION_RUN_URL }})"