环境配置

主机名	IP	k8s版本	备注	操作系统	nvidia-smi版本	cuda版本	显卡
master1	192.168.10.32	v1.28.0	master1	22.04 LTS	550.120	12.4	NVIDIA A10
node1	192.168.10.33	v1.28.0	node1	22.04 LTS	550.120	12.4	NVIDIA A10

一.安装显卡驱动

1.查看显卡

root@gpu:~# apt update
root@gpu:~# lspci | grep -i nvidia
00:07.0 3D controller: NVIDIA Corporation GA102GL [A10] (rev a1)

2.安装驱动

#推荐我们使用550驱动
root@gpu:~# ubuntu-drivers devices
model    : GA102GL [A10]
driver   : nvidia-driver-550 - distro non-free recommended

#安装550驱动
apt install -y nvidia-driver-550

3.查看安装好的驱动

root@gpu:~# nvidia-smi
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.120                Driver Version: 550.120        CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA A10                     Off |   00000000:00:07.0 Off |                    0 |
|  0%   40C    P0             5W /  150W |       1MiB /  23028MiB |     0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

二.安装cuda toolkit

因为安装的nvidia版本是550，所以我们安装cuda是对应的12.4

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
dpkg -i cuda-keyring_1.1-1_all.deb
apt-get update
apt-get -y install cuda-toolkit-12-4

root@gpu:~# vim /etc/profile
export PATH=$PATH:/usr/local/cuda-12.4/bin/
root@gpu:~# source /etc/profile

三.重启

reboot

四.安装k8s

1.配置主机名

root@master1:~# hostnamectl set-hostname master1
root@node1:~# hostnamectl set-hostname node1

2.配置host文件

root@master1:~# cat >> /etc/hosts << EOF
192.168.10.32 master1
192.168.10.33 node1
EOF

3.系统优化

#转发 IPv4 并让 iptables 看到桥接流量
cat <<EOF | sudo tee /etc/modules-load.d/k8s.conf
overlay
br_netfilter
EOF

sudo modprobe overlay
sudo modprobe br_netfilter

# 设置所需的 sysctl 参数，参数在重新启动后保持不变
cat <<EOF | sudo tee /etc/sysctl.d/k8s.conf
net.bridge.bridge-nf-call-iptables  = 1
net.bridge.bridge-nf-call-ip6tables = 1
net.ipv4.ip_forward                 = 1
EOF

# 应用 sysctl 参数而不重新启动
sudo sysctl --system

# 临时关闭；关闭swap主要是为了性能考虑
swapoff -a

# 永久关闭
sed -ri 's/.*swap.*/#&/' /etc/fstab

4.安装containerd

4.1安装nvidia-container-toolkit

curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
  && curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
    sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
    sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

sed -i -e '/experimental/ s/^#//g' /etc/apt/sources.list.d/nvidia-container-toolkit.list

apt-get update
apt-get install -y nvidia-container-toolkit

4.2安装containerd

apt update
apt install -y containerd

# 生成containetd的配置文件
mkdir -p /etc/containerd/
containerd config default | sudo tee /etc/containerd/config.toml >/dev/null 2>&1

#使用nvidia-ctk配置容器运行时为containerd
nvidia-ctk runtime configure --runtime=containerd

# 修改/etc/containerd/config.toml，修改SystemdCgroup为true
sed -i "s#SystemdCgroup\ \=\ false#SystemdCgroup\ \=\ true#g" /etc/containerd/config.toml
cat /etc/containerd/config.toml | grep SystemdCgroup

# 修改沙箱镜像源
sed -i "s#registry.k8s.io/pause#registry.cn-hangzhou.aliyuncs.com/google_containers/pause#g" /etc/containerd/config.toml
cat /etc/containerd/config.toml | grep sandbox_image

#修改containerd镜像拉取
vim /etc/containerd/config.toml
    [plugins."io.containerd.grpc.v1.cri".registry]
      config_path = ""

      [plugins."io.containerd.grpc.v1.cri".registry.mirrors."docker.io"]
       endpoint = ["https://teut9a0h.mirror.aliyuncs.com"]
      [plugins."io.containerd.grpc.v1.cri".registry.mirrors."k8s.gcr.io"]
       endpoint = ["registry.aliyuncs.com/google_containers"]
       
修改runtime为nvidia
[plugins."io.containerd.grpc.v1.cri".containerd]
  default_runtime_name = "nvidia"
       
root@master1:~# systemctl restart containerd

4.3安装crictl

#添加 apt key
curl https://mirrors.aliyun.com/kubernetes/apt/doc/apt-key.gpg | sudo apt-key add -

#添加Kubernetes的apt源为阿里云的源
apt-add-repository "deb https://mirrors.aliyun.com/kubernetes/apt/ kubernetes-xenial main"

# 安装crictl工具
apt -y install cri-tools

# 生成配置文件
crictl config runtime-endpoint

# 编辑配置文件
cat << EOF | tee /etc/crictl.yaml
runtime-endpoint: "unix:///run/containerd/containerd.sock"
image-endpoint: "unix:///run/containerd/containerd.sock"
timeout: 10
debug: false
pull-image-on-create: false
disable-pull-on-run: false
EOF

5.安装k8s组件

apt install -y kubelet=1.28.0-00 kubeadm=1.28.0-00 kubectl=1.28.0-00
systemctl enable kubelet && systemctl start kubelet

6.初始化master(master1节点操作)

#查看镜像
kubeadm  config images list --kubernetes-version=v1.28.0 --image-repository=registry.aliyuncs.com/google_containers

#下载镜像
kubeadm  config images pull --kubernetes-version=v1.28.0 --image-repository=registry.aliyuncs.com/google_containers

#初始化k8s
kubeadm init \
  --apiserver-advertise-address=192.168.10.32 \
  --image-repository registry.aliyuncs.com/google_containers \
  --kubernetes-version v1.28.0 \
  --service-cidr=10.96.0.0/12 \
  --pod-network-cidr=10.244.0.0/16 \
  --ignore-preflight-errors=all

拷贝kubectl使用的k8s认证文件到默认路径

mkdir -p $HOME/.kube
sudo cp -i /etc/kubernetes/admin.conf $HOME/.kube/config
sudo chown $(id -u):$(id -g) $HOME/.kube/config
  
export KUBECONFIG=/etc/kubernetes/admin.conf

7.node节点加入集群(node1节点操作)

kubeadm join 192.168.10.32:6443 --token 7hwuxh.bvfsd9kia04nx7qq \
        --discovery-token-ca-cert-hash sha256:190a72151f58f3632633ca377ae8d2f1517d5bbf401f390a3be67fd091185ac3        

8.安装flannel网络插件

wget https://raw.githubusercontent.com/coreos/flannel/master/Documentation/kube-flannel.yml

kubectl apply -f kube-flannel.yml

9.查看所有节点

root@master1:~# kubectl get node
NAME      STATUS   ROLES           AGE   VERSION
master1   Ready    control-plane   15m   v1.28.0
node1     Ready    <none>          15m   v1.28.0

五.安装nvidia-device-plugin插件

kubectl apply -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.14.1/nvidia-device-plugin.yml

root@master1:~# kubectl describe node master1| grep nvidia.com/gpu
  nvidia.com/gpu:     1
  nvidia.com/gpu:     1
  nvidia.com/gpu     0           0
root@master1:~# kubectl describe node node1| grep nvidia.com/gpu
  nvidia.com/gpu:     1
  nvidia.com/gpu:     1
  nvidia.com/gpu     0           0

六.部署deepseek

1.下载镜像

#官方镜像
vllm/vllm-openai:latest
ghcr.io/open-webui/open-webui:cuda

#自建阿里云镜像仓库
registry.cn-beijing.aliyuncs.com/tools_cluster/tools_cluster:vllm-openai
registry.cn-beijing.aliyuncs.com/tools_cluster/tools_cluster:open-webui-cuda

crictl pull registry.cn-beijing.aliyuncs.com/tools_cluster/tools_cluster:vllm-openai
crictl pull registry.cn-beijing.aliyuncs.com/tools_cluster/tools_cluster:open-webui-cuda

2.下载deepseek-R1-1.5B模型(二选一)

2.1modelscope

https://modelscope.cn/home

mkdir /root/model
pip install modelscope setuptools
modelscope download --model deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B --local_dir /root/model

2.2huggingface

https://huggingface.co/

pip install huggingface_hub
mkdir /root/model
export HF_ENDPOINT=https://hf-mirror.com
huggingface-cli download --resume-download deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B --local-dir /root/model

3.部署deepseek

kubectl create ns deepseek
vim deepseek.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: deepseek-vllm
  namespace: deepseek
spec:
  replicas: 1
  selector:
    matchLabels:
      app: deepseek-vllm
  template:
    metadata:
      labels:
        app: deepseek-vllm
    spec:
      nodeName: node1
      containers:
      - name: vllm
        image: registry.cn-beijing.aliyuncs.com/tools_cluster/tools_cluster:vllm-openai
        args: [
          "--served-model-name", "DeepSeek-R1:1.5B",
          "--trust-remote-code",
          "--enforce-eager",
          "--enable-chunked-prefill",
          "--gpu-memory-utilization", "0.95",
          "--model", "/app/model/",
          "--host", "0.0.0.0",
          "--port", "8000",
          "--api-key", "12345678",
          "--tokenizer", "/app/model/",
          "--tensor-parallel-size", "1"
        ]
        ports:
        - containerPort: 8000
        resources:
          requests:
            nvidia.com/gpu: "1"
          limits:
            nvidia.com/gpu: 1
        volumeMounts:
        - name: data
          mountPath: /app/model
      volumes:
      - name: data
        hostPath:
          path: /root/model
---
apiVersion: v1
kind: Service
metadata:
  name: vllm-api
  namespace: deepseek
spec:
  selector:
    app: vllm
  type: ClusterIP
  ports:
  - name: api
    protocol: TCP
    port: 8000
    targetPort: 8000

kubectl apply -f deepseek.yaml

4.部署webui

apiVersion: apps/v1
kind: Deployment
metadata:
  name: webui
  namespace: deepseek
spec:
  replicas: 1
  selector:
    matchLabels:
      app: webui
  template:
    metadata:
      labels:
        app: webui
    spec:
      nodeName: master1
      containers:
      - name: webui
        image: registry.cn-beijing.aliyuncs.com/tools_cluster/tools_cluster:open-webui-cuda
        env:
        - name: OPENAI_API_BASE_URL
          value: http://vllm-api:8000/v1
        - name: ENABLE_OLLAMA_API
          value: "False"
        - name: OPENAI_API_KEYS
          value: "12345678"
        - name: USE_CUDA_DOCKER
          value: "true"
        ports:
        - containerPort: 8080

---
apiVersion: v1
kind: Service
metadata:
  name: webui
  namespace: deepseek
  labels:
    app: webui
spec:
  type: ClusterIP
  ports:
  - port: 8080
    protocol: TCP
    targetPort: 8080
  selector:
    app: webui

5.访问接口测试

curl --location 'http://10.244.1.27:8000/v1/chat/completions' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer 12345678' \
--data '{
    "model":"DeepSeek-R1:1.5B",
    "messages": [
      {
        "role": "user",
        "content": "你是谁？"
      }
    ]
}'