add gpu node config

.gitignore

@@ -6,3 +6,4 @@ hosts
 *.crt
 *.pem
 roles/prepare/files/ca*
+.idea

21.gpunode.yml

@@ -0,0 +1,3 @@
+- hosts: gpu-node
+  roles:
+  - gpu

docs/08-配置GPU-node节点.md

@@ -0,0 +1,46 @@
+## 08-配置GPU-node节点.md
+
+推荐阅读[官方GPU节点配置文档](https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/)
+
+### 1. 允许k8s系统使用Device Plugins
+
+实现方式：通过修改kube-apiserver, kubelet, kube-proxy配置文件模板
+
+因为GPU node是kube node的子集，正常执行`90.setup.yml`搭建k8s即可实现此目标
+
+### 2. 配置GPU节点
+在[官方驱动网页](http://www.nvidia.com/Download/index.aspx?lang=en-uk)下载对应操作系统与显卡的驱动，
+改名为nvidia-diag-driver-local-repo.deb，
+并放入到invertory file中{{ base_dir }}路线下的 bin 文件夹中
+
+执行命令：`ansible-playbook 21.gpunode.yml`
+
+相当于完成以下任务
+
+#### 1). GPU 节点安装 nvidia driver 
+建议在node上只安装驱动，而不安装CUDA包，所有CUDA包都放到镜像中去，否则容易出现版本不匹配的问题。
+
+
+
+#### 2). GPU 节点安装 nvidia docker 2.0
+这一部分可能失败，原因可能是系统所用docker版本与nvidia docker 2.0依赖的docker版本不一致，
+具体参考[文档](https://github.com/NVIDIA/nvidia-docker/wiki/Frequently-Asked-Questions)
+
+
+可以通过命令`apt-cache madison nvidia-docker2 nvidia-container-runtime`查询可以安装的nvidia docker 2.0 
+的 docker 版本
+
+可以通过命令`dpkg -l | grep docker`来查看实际安装docker 版本
+
+nvidia docker 安装失败，可以[手动更新docker](https://docs.docker.com/engine/installation/linux/docker-ce/ubuntu/#upgrade-docker-ce),
+再重新运行安装playbook
+
+
+#### 3). GPU 节点配置 nvidia-container-runtime 为 docker default runtime
+
+通过修改`/etc/docker/daemon.json`实现
+
+### 3. 配置Nvidia device plugin
+执行命令 `kubectl create -f manifests/gpu-device-plugin/v1.9/nvidia-device-plugin.yml`
+
+可以通过执行`kubectl describe nodes | grep nvidia.com/gpu` 来查看GPU节点配置是否成功

example/hosts.allinone.example

@@ -14,6 +14,11 @@
 [kube-node]
 192.168.1.1 NODE_ID=node1 NODE_IP="192.168.1.1"
 
+#gpu-node 是 kube-node的子集
+[gpu-node]
+192.168.1.1 NODE_IP="192.168.1.1"
+
+
 [kube-cluster:children]
 kube-node
 kube-master

example/hosts.m-masters.example

@@ -30,6 +30,12 @@ MASTER_PORT="8443"		# api-server 服务端口
 192.168.1.3 NODE_ID=node2 NODE_IP="192.168.1.3"
 192.168.1.4 NODE_ID=node3 NODE_IP="192.168.1.4"
 
+#gpu-node 是 kube-node的子集
+[gpu-node]
+192.168.1.2 NODE_IP="192.168.1.2"
+192.168.1.3 NODE_IP="192.168.1.3"
+
+
 [kube-cluster:children]
 kube-node
 kube-master

example/hosts.s-master.example

@@ -18,6 +18,12 @@
 192.168.1.2 NODE_ID=node2 NODE_IP="192.168.1.2"
 192.168.1.3 NODE_ID=node3 NODE_IP="192.168.1.3"
 
+#gpu-node 是 kube-node的子集
+[gpu-node]
+192.168.1.2 NODE_IP="192.168.1.2"
+192.168.1.3 NODE_IP="192.168.1.3"
+
+
 [kube-cluster:children]
 kube-node
 kube-master

manifests/gpu-device-plugin/v1.8/nvidia-device-plugin.yml

@@ -0,0 +1,24 @@
+apiVersion: extensions/v1beta1
+kind: DaemonSet
+metadata:
+  name: nvidia-device-plugin-daemonset
+spec:
+  template:
+    metadata:
+      labels:
+        name: nvidia-device-plugin-ds
+    spec:
+      containers:
+      - image: nvidia/k8s-device-plugin:1.8
+        name: nvidia-device-plugin-ctr
+        securityContext:
+          allowPrivilegeEscalation: false
+          capabilities:
+            drop: ["ALL"]
+        volumeMounts:
+          - name: device-plugin
+            mountPath: /var/lib/kubelet/device-plugins
+      volumes:
+        - name: device-plugin
+          hostPath:
+            path: /var/lib/kubelet/device-plugins

manifests/gpu-device-plugin/v1.9/nvidia-device-plugin.yml

@@ -0,0 +1,25 @@
+apiVersion: extensions/v1beta1
+kind: DaemonSet
+metadata:
+  name: nvidia-device-plugin-daemonset
+  namespace: kube-system
+spec:
+  template:
+    metadata:
+      labels:
+        name: nvidia-device-plugin-ds
+    spec:
+      containers:
+      - image: nvidia/k8s-device-plugin:1.9
+        name: nvidia-device-plugin-ctr
+        securityContext:
+          allowPrivilegeEscalation: false
+          capabilities:
+            drop: ["ALL"]
+        volumeMounts:
+          - name: device-plugin
+            mountPath: /var/lib/kubelet/device-plugins
+      volumes:
+        - name: device-plugin
+          hostPath:
+            path: /var/lib/kubelet/device-plugins

roles/gpu/files/daemon.json

@@ -0,0 +1,12 @@
+{
+  "registry-mirrors": ["https://registry.docker-cn.com"], 
+  "max-concurrent-downloads": 6,
+  "default-runtime": "nvidia",
+  "runtimes": {
+      "nvidia": {
+          "path": "/usr/bin/nvidia-container-runtime",
+          "runtimeArgs": []
+      }
+  }
+}
+

roles/gpu/tasks/main.yml

@@ -0,0 +1,30 @@
+- name: 下载nvidia driver安装包
+  copy: src={{ base_dir }}/bin/nvidia-diag-driver-local-repo.deb dest=/tmp/nvidia-diag-driver-local-repo.deb
+  when: ansible_distribution == "Ubuntu"
+
+- name: 安装nvidia dirver驱动
+  apt:
+    deb: /tmp/nvidia-diag-driver-local-repo.deb
+  when: ansible_distribution == "Ubuntu"
+
+- name: Add the package repositories 1
+  shell: curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | apt-key add -
+  when: ansible_distribution == "Ubuntu"
+
+- name: Add the package repositories 2
+  shell: curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu16.04/amd64/nvidia-docker.list | tee /etc/apt/sources.list.d/nvidia-docker.list
+  when: ansible_distribution == "Ubuntu"
+
+- name: Update apt-get and Install nvidia docker 2
+  apt:
+    name: nvidia-docker2
+    update_cache: yes
+  when: ansible_distribution == "Ubuntu"
+
+- name: 配置nvidia docker runtime
+  copy: src=daemon.json dest=/etc/docker/daemon.json
+
+- name: restart docker
+  service:
+    name: docker
+    state: restarted

roles/kube-master/templates/kube-apiserver.service.j2

@@ -31,6 +31,7 @@ ExecStart={{ bin_dir }}/kube-apiserver \
   --audit-log-maxsize=100 \
   --audit-log-path=/var/lib/audit.log \
   --event-ttl=1h \
+  --feature-gates="DevicePlugins=true" \
   --v=2
 Restart=on-failure
 RestartSec=5

roles/kube-node/templates/kube-proxy.service.j2

@@ -12,6 +12,7 @@ ExecStart={{ bin_dir }}/kube-proxy \
   --hostname-override={{ NODE_IP }} \
   --kubeconfig=/etc/kubernetes/kube-proxy.kubeconfig \
   --logtostderr=true \
+  --feature-gates="DevicePlugins=true" \
   --v=2
 Restart=on-failure
 RestartSec=5

roles/kube-node/templates/kubelet.service.j2

@@ -23,6 +23,7 @@ ExecStart={{ bin_dir }}/kubelet \
   --allow-privileged=true \
   --fail-swap-on=false \
   --logtostderr=true \
+  --feature-gates="DevicePlugins=true" \
   --v=2
 #kubelet cAdvisor 默认在所有接口监听 4194 端口的请求, 以下iptables限制内网访问
 ExecStartPost=/sbin/iptables -A INPUT -s 10.0.0.0/8 -p tcp --dport 4194 -j ACCEPT