diff --git a/.github/scripts/build_aiter_triton.sh b/.github/scripts/build_aiter_triton.sh index 5ad68cbffa..05b60f7ca5 100755 --- a/.github/scripts/build_aiter_triton.sh +++ b/.github/scripts/build_aiter_triton.sh @@ -25,7 +25,7 @@ if [[ "$BUILD_TRITON" == "1" ]]; then cd triton pip install -r python/requirements.txt pip install filecheck - MAX_JOBS=64 pip --retries=5 install . + MAX_JOBS=64 pip --retries=10 --default-timeout=60 install . cd .. else echo diff --git a/.github/workflows/aiter-test.yaml b/.github/workflows/aiter-test.yaml index a650c5a1f2..8973c509d3 100644 --- a/.github/workflows/aiter-test.yaml +++ b/.github/workflows/aiter-test.yaml @@ -104,6 +104,11 @@ jobs: --name aiter_test \ ${{ env.DOCKER_IMAGE }} + - name: Setup pip config + run: | + docker exec -u root aiter_test bash -c "pip config set global.default-timeout 60" + docker exec -u root aiter_test bash -c "pip config set global.retries 10" + - name: Setup Aiter run: | set -ex @@ -183,10 +188,15 @@ jobs: --name aiter_test \ ${{ env.DOCKER_IMAGE }} - - name: Setup-Triton + - name: Setup pip config + run: | + docker exec -u root aiter_test bash -c "pip config set global.default-timeout 60" + docker exec -u root aiter_test bash -c "pip config set global.retries 10" + + - name: Setup-Aiter run: | set -ex - echo "Setting up Triton..." + echo "Setting up Aiter..." docker exec \ -w /workspace \ aiter_test \ diff --git a/.github/workflows/sglang_downstream.yaml b/.github/workflows/sglang_downstream.yaml index 3c66e3b6ec..58421cda00 100644 --- a/.github/workflows/sglang_downstream.yaml +++ b/.github/workflows/sglang_downstream.yaml @@ -99,6 +99,11 @@ jobs: env: GITHUB_WORKSPACE: ${{ github.workspace }} + - name: Setup pip config + run: | + docker exec -u root sglang_aiter_test bash -c "pip config set global.default-timeout 60" + docker exec -u root sglang_aiter_test bash -c "pip config set global.retries 10" + - name: Install dependencies run: | cd sglang diff --git a/.github/workflows/test-network.yaml b/.github/workflows/test-network.yaml new file mode 100644 index 0000000000..bc52339a49 --- /dev/null +++ b/.github/workflows/test-network.yaml @@ -0,0 +1,78 @@ +name: Test Connection to PyPI and GitHub + +on: + workflow_dispatch: + inputs: + runner: + description: 'Runner to use for the test, e.g. aiter-mi355-1gpu, aiter-1gpu-runner, etc.' + required: true + default: 'aiter-mi355-1gpu' + +concurrency: + group: ${{ github.workflow }}-${{ github.ref }} + cancel-in-progress: true + +jobs: + test-pypi: + runs-on: aiter-mi355-1gpu + steps: + - name: Checkout code + uses: actions/checkout@v4 + + - name: Run the container + run: | + set -ex + echo "Starting container: test_pypi" + docker run -dt \ + -v "${{ github.workspace }}:/workspace" \ + -w /workspace \ + --name test_pypi \ + python:3.10-slim \ + + - name: Test network speed and latency + run: | + set -ex + # ==== Network speed test with speedtest-cli ==== + echo -e "\033[1;34m==== Network Speed Test (speedtest-cli) ====\033[0m" + + docker exec -u root test_pypi bash -c "pip config set global.default-timeout 60" + docker exec -u root test_pypi bash -c "pip config set global.retries 10" + + # Install and show speedtest-cli results + docker exec \ + -w /workspace \ + test_pypi \ + bash -c "pip install -q speedtest-cli && echo -e '\033[1;33mSpeedtest-cli quick summary:\033[0m' && speedtest-cli --simple || true" + + # ==== DNS and latency test for pypi.org and github.com ==== + echo -e "\033[1;34m==== Checking DNS and Latency for pypi.org and github.com ====\033[0m" + docker exec \ + -w /workspace \ + test_pypi \ + bash -c "apt-get update && apt-get install -y dnsutils iputils-ping curl" + + # DNS lookup for pypi.org + echo -e "\033[1;36m[pypi.org] DNS lookup:\033[0m" + docker exec test_pypi bash -c "nslookup pypi.org || true" + + # DNS lookup for github.com + echo -e "\033[1;36m[github.com] DNS lookup:\033[0m" + docker exec test_pypi bash -c "nslookup github.com || true" + + # Ping test for pypi.org and github.com + echo -e "\033[1;36m[pypi.org] Ping latency:\033[0m" + docker exec test_pypi bash -c "ping -c 4 pypi.org || true" + echo -e "\033[1;36m[github.com] Ping latency:\033[0m" + docker exec test_pypi bash -c "ping -c 4 github.com || true" + + # ==== Download speed test for pypi.org ==== + echo -e "\033[1;34m==== Testing Download Speed from pypi.org ====\033[0m" + docker exec \ + test_pypi \ + bash -c "curl -o /dev/null -L --max-time 60 https://files.pythonhosted.org/packages/source/p/pip/pip-24.0.tar.gz -w 'pypi.org download: %{speed_download} bytes/sec\n'" + + # ==== Download speed test for github.com ==== + echo -e "\033[1;34m==== Testing Download Speed from github.com ====\033[0m" + docker exec \ + test_pypi \ + bash -c "curl -o /dev/null -L --max-time 60 https://github.com/git/git/archive/refs/tags/v2.42.0.tar.gz -w 'github.com download: %{speed_download} bytes/sec\n'" diff --git a/.github/workflows/triton-test.yaml b/.github/workflows/triton-test.yaml index 8f2aff5cd5..4c622e7bd4 100644 --- a/.github/workflows/triton-test.yaml +++ b/.github/workflows/triton-test.yaml @@ -69,6 +69,11 @@ jobs: triton_test \ bash -c "pip install speedtest-cli && speedtest-cli --simple" || true + - name: Setup pip config + run: | + docker exec -u root triton_test bash -c "pip config set global.default-timeout 60" + docker exec -u root triton_test bash -c "pip config set global.retries 10" + - name: Setup Aiter and Triton run: | set -ex diff --git a/.github/workflows/vllm_benchmark.yaml b/.github/workflows/vllm_benchmark.yaml index d800d208f5..358f563eb8 100644 --- a/.github/workflows/vllm_benchmark.yaml +++ b/.github/workflows/vllm_benchmark.yaml @@ -60,6 +60,8 @@ jobs: RUN echo "=== Aiter version BEFORE uninstall ===" && pip show aiter || true RUN pip uninstall -y aiter + RUN pip config set global.default-timeout 60 + RUN pip config set global.retries 10 RUN pip install --upgrade "pybind11>=3.0.1" RUN pip show pybind11