feat: metric e2e 启动服务中判断是否当前服务器存在可用GPU,如无则跳过 test-gpu-node 容器创建和安装包的测试;

refs #29
This commit is contained in:
sundapeng.sdp
2025-10-20 15:30:07 +08:00
parent 299765ed40
commit c4582c99bc
3 changed files with 116 additions and 14 deletions

View File

@@ -0,0 +1,59 @@
#!/bin/bash
# GPU环境检测脚本
# 检测系统是否有NVIDIA GPU硬件
set -e
# 检测函数
check_gpu_support() {
echo "检测GPU环境..."
# 方法1: 检测GPU设备文件
if ls /dev/nvidia* &>/dev/null; then
echo "✓ 检测到NVIDIA GPU设备文件"
return 0
fi
# 方法2: 检测lspci中的NVIDIA设备(Linux)
if command -v lspci &> /dev/null; then
if lspci | grep -i nvidia &> /dev/null; then
echo "✓ 检测到NVIDIA GPU硬件"
return 0
fi
fi
# 方法3: 检测nvidia-smi
if command -v nvidia-smi &> /dev/null; then
if nvidia-smi &> /dev/null; then
echo "✓ 检测到NVIDIA GPU硬件"
return 0
fi
fi
echo "✗ 未检测到NVIDIA GPU硬件"
return 1
}
# 主函数
main() {
echo "=========================================="
echo " GPU环境检测"
echo "=========================================="
echo ""
if check_gpu_support; then
echo ""
echo "结果: GPU环境可用"
exit 0
else
echo ""
echo "结果: GPU环境不可用,将跳过GPU相关服务"
exit 1
fi
}
# 如果直接运行此脚本
if [ "${BASH_SOURCE[0]}" = "${0}" ]; then
main "$@"
fi