k8s 调度 GPU

最近公司有项目想在 k8s 集群中运行 GPU 任务，于是研究了一下。下面是部署的步骤。

1. 首先得有一个可以运行的 k8s 集群. 集群部署参考 kubeadm安装k8s 

2. 准备 GPU 节点

2.1 安装驱动

curl -fsSL https://mirrors.aliyun.com/nvidia-cuda/ubuntu1804/x86_64/7fa2af80.pub | sudo apt-key add -
echo "deb https://mirrors.aliyun.com/nvidia-cuda/ubuntu1804/x86_64/ ./" > /etc/apt/sources.list.d/cuda.list

apt-get update
apt-get install -y cuda-drivers-455 # 按需要安装对应的版本

2.2 安装 nvidia-docker2

<!– Note that you need to install the nvidia-docker2 package and not the nvidia-container-toolkit. This is because the new –gpus options hasn’t reached kubernetes yet –>

distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list

sudo apt-get update && sudo apt-get install -y nvidia-docker2

## /etc/docker/daemon.json 文件中加入以下内容, 使默认的运行时是 nvidia
{
    "default-runtime": "nvidia",
    "runtimes": {
        "nvidia": {
            "path": "/usr/bin/nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}

## 重启 docker
sudo systemctl restart docker

2.3 在 k8s 集群中安装 nvidia-device-plugin 使集群支持 GPU

kubectl create -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.7.3/nvidia-device-plugin.yml

# 如果因为网络问题访问不到该文件, 可在浏览器打开 https://github.com/NVIDIA/k8s-device-plugin/blob/v0.7.3/nvidia-device-plugin.yml
## 把文件内容拷贝到本地执行

    nvidia-device-plugin 做三件事情

• Expose the number of GPUs on each nodes of your cluster

• Keep track of the health of your GPUs

• Run GPU enabled containers in your Kubernetes cluster.

之后把节点加入 k8s 集群

以上步骤成功完成之后, 运行以下命令能看到类似下面图片中的内容说明插件安装好了

kubectl get pod --all-namespaces | grep nvidia
kubectl describe node 10.31.0.17

3. 运行 GPU Jobs

# cat nvidia-gpu-demo.yaml
apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
spec:
  containers:
    - name: cuda-container
      image: nvidia/cuda:9.0-devel
      resources:
        limits:
          nvidia.com/gpu: 2 # requesting 2 GPUs
    - name: digits-container
      image: nvidia/digits:6.0
      resources:
        limits:
          nvidia.com/gpu: 2 # requesting 2 GPUs

kubectl apply -f nvidia-gpu-demo.yaml

kubectl exec -it xxx-76dd5bd849-hlmdr -- bash

# nvidia-smi

以上就简单实现了 k8s 调度 GPU 任务。 

如有遇到问题可在留言区讨论。