MLU-K8S-PLUGIN安装

1 安装过程

1. 1.  环境要求

MLU100, MLU270, x5k, MLU220 devices

MLU100 driver > 3.5; MLU270 driver >2.2.0; MLU220 driver > 4.1.1

libcndev.so >= V1.8.0

Kubernetes >= v1.11.2

1. 1.  下载与构建

（以下步骤需在联网环境进行，如离线安装，需在通网节点下载）

1.2.1 Clone 包：

git clone GitHub - Cambricon/cambricon-k8s-device-plugin

1.2.2 进入安装目录：

cd cambricon-k8s-device-plugin/device-plugin

1.2.3 构建镜像：

    如离线安装，需修改build_image.sh 以及Dockerfile，修改内容见附录；

        ./build_image.sh

        注意：构建完成后将镜像传送至安装节点；

确保Cambricon neuware已安装，在构建镜像过程中会需要libcndev.so。

1. 1.  加载镜像

docker load -i image/cambricon-k8s-device-plugin-amd64.tar

1. 1. 部署守护进程

1.4.1修改yaml文件

vim ./example/cambricon-device-plugin-daemonset.yaml

可修改参数：

args:

  - --mode=default #device plugin mode: default, sriov or env-share

      - --virtualization-num=1 # virtualization number for each MLU, used only in sriov mode or env-share mode

1.4.2启动进程

kubectl create -f cambricon-device-plugin-daemonset.yaml

1. 1. 使用MLU运行任务

Cambricon MLU现在可以通过容器级资源需求使用，使用资源名称：‎cambricon.com/mlu

例：

apiVersion: v1

kind: Pod

metadata:

  name: pod1

spec:

  restartPolicy: OnFailure

  containers:

    - image: ubuntu:16.04

      name: pod1-ctr

      command: ["sleep"]

      args: ["100000"]

      resources:

        limits:

          cambricon.com/mlu: 1

1. 问题解决
   1. 构建镜像过程中出现以下问题：

Err:1 http://deb.debian.org/debian buster InRelease                 

  Temporary failure resolving 'deb.debian.org'

Err:2 http://security.debian.org/debian-security buster/updates InRelease

  Temporary failure resolving 'security.debian.org'

Err:3 http://deb.debian.org/debian buster-updates InRelease         

  Temporary failure resolving 'deb.debian.org'

Reading package lists... Done   

W: Failed to fetch http://deb.debian.org/debian/dists/buster/InRelease  Temporary failure resolving 'deb.debian.org'

W: Failed to fetch http://security.debian.org/debian-security/dists/buster/updates/InRelease  Temporary failure resolving 'security.debian.org'

W: Failed to fetch http://deb.debian.org/debian/dists/buster-updates/InRelease  Temporary failure resolving 'deb.debian.org'

W: Some index files failed to download. They have been ignored, or old ones used instead.

解决办法：

vim /etc/docker/daemon.json，

添加行 "dns": ["114.114.114.114","8.8.8.8"]

重启docker：systemctl restart docker

1. 1. 报错：file not found in build context or excluded by .dockerignore

  原因：dockerfile 不能获取 父目录

  解决办法：将文件copy到当前目录

1. 1. Pause k8s 镜像下载失败

如果kubernetes集群在内网环境中，无法访问gcr.io网站，则可先通过一台能访问gcr.io的机器下载pause镜像，导出后再导入内网的docker私有镜像仓库中，并在kubelet的启动参数中加上--pod_infra_container_image，然后重启kubelet.

docker pull kubernetes/pause

1. 1. spec.template.spec.containers[0].securityContext.privileged:Forbidden: disallowed by policy问题

解决方法：kube-apiserver和kubelet的启动脚本中添加--allow_privileged=true

步骤：

1.管理节点vim /etc/sysconfig/kube-apiserver

2. 修改KUBE_APISERVER_OPTS='--allow_privileged=true'

3. systemctl daemon-reload

  systemctl restart kube-apiserver

  systemctl status -l kube-apiserver

4.计算节点 vi /etc/sysconfig/kubelet

5. 修改KUBELET_OPTS='--allow_privileged=true'

6. systemctl daemon-reload

  systemctl restart kubelet

  systemctl status -l kubelet

1. 1. http: server gave HTTP response to HTTPS client

解决方法：vim /etc/docker/daemon.json

         修改{ "insecure-registries":["xxxxxxxxx:5000"] }

        systemctl daemon-reload

        systemctl restart docker