docker-compose编排动态配置的pbsPro容器集群

简介

docker-compose是一种容器编排工具，可以将多个docker容器关联部署。通过yaml文件，可以描述应用的架构，如使用什么镜像、数据卷、网络、绑定服务端口等等，然后再用一条命令就可以管理所有的服务（如启动、停止、重启、日志监控等等），可以鼎泰扩展容器的规模等等，和docker swarm非常相似又有区别。
本次博客将对前一节利用Dockerfile编排pbspro集群进行升级，使用docker-compose来将pbsPro集群部署起来，结合shell脚本，可以动态配置整个集群的编排。所有脚本已上传至github上。

---

编排文件脚本下载

1. 构建基础镜像maya-pbsPro

从前一步的zxj/maya-pbspro：latest镜像开始,该镜像的获取可参考前面镜像制作文章：利用Dockerfile编排pbspro集群，或者拉取本人的镜像库

$# docker pull zxj2017/maya-pbspro

1. 构建新镜像zxj/maya-pbs:onbuild

---

$# cd centos-maya-openpbs/onbuild

#copy configutation file and ssh
FROM zxj/maya-pbs:latest
MAINTAINER zxj
USER root

RUN yum install openssh-clients \
    yum install openssh-server

# # ------------------------------------------------------------
# # Utility shell scripts
# # ------------------------------------------------------------

COPY mpi_bootstrap /usr/local/bin/mpi_bootstrap
RUN chmod +x /usr/local/bin/mpi_bootstrap

COPY get_hosts /usr/local/bin/get_hosts
RUN chmod +x /usr/local/bin/get_hosts

COPY auto_update_hosts /usr/local/bin/auto_update_hosts
RUN chmod +x /usr/local/bin/auto_update_hosts

# # ------------------------------------------------------------
# # Miscellaneous setup for better user experience
# # ------------------------------------------------------------

# Set welcome message to display when user ssh login 
COPY welcome.txt /etc/motd

# Default hostfile location for mpirun. This file will be updated automatically.
ENV HYDRA_HOST_FILE /etc/opt/hosts
RUN echo "export HYDRA_HOST_FILE=${HYDRA_HOST_FILE}" >> /etc/profile

RUN touch ${HYDRA_HOST_FILE}
RUN chown ${USER}:${USER} ${HYDRA_HOST_FILE}

# Auto go to default working directory when user ssh login 
RUN echo "cd $WORKDIR" >> ${USER_HOME}/.profile

# # ------------------------------------------------------------
# # Set up SSH Server 
# # ------------------------------------------------------------

# Add host keys
RUN cd /etc/ssh/ && ssh-keygen -A -N ''

# Config SSH Daemon
RUN  sed -i "s/#PasswordAuthentication.*/PasswordAuthentication no/g" /etc/ssh/sshd_config \
  && sed -i "s/#PermitRootLogin.*/PermitRootLogin no/g" /etc/ssh/sshd_config \
  && sed -i "s/#AuthorizedKeysFile/AuthorizedKeysFile/g" /etc/ssh/sshd_config

# Unlock non-password USER to enable SSH login
RUN echo "123456" | passwd --stdin "root"
#passwd -u ${USER}

# Set up user's public and private keys
ENV SSHDIR ${USER_HOME}/.ssh
RUN mkdir -p ${SSHDIR}

# Default ssh config file that skips (yes/no) question when first login to the host
RUN echo "StrictHostKeyChecking no" > ${SSHDIR}/config
# This file can be overwritten by the following onbuild step if ssh/ directory has config file

# Switch back to default user
USER ${USER}

# # ------------------------------------------------------------
# # ONBUILD (require ssh/ directory in the build context)
# # ------------------------------------------------------------
ONBUILD USER root

ONBUILD COPY ssh/ ${SSHDIR}/

ONBUILD RUN cat ${SSHDIR}/*.pub >> ${SSHDIR}/authorized_keys
ONBUILD RUN chmod -R 600 ${SSHDIR}/* \
         && chown -R ${USER}:${USER} ${SSHDIR}

# Switch back to default user when continue the build process
ONBUILD USER ${USER}

使用编译命令，构建成为一个zxj/maya-pbs:onbuild镜像

---

$#   docker build -t zxj/maya-pbs:onbuild .

---

---

2 .构建docker-compose.yml文件

由上一步构建的镜像zxj/maya-pbs:onbuild，后续继续使用docker compose在这个镜像上编排，编排文件如下

---

cd centos-maya-openpbs/cluster

version: "2"

services:
  registry:
    image: registry
    ports:
      - "${REGISTRY_PORT}:5000"

  master:
    image: $REGISTRY_ADDR:$REGISTRY_PORT/$IMAGE_NAME
    user: root
    hostname: master
    privileged: true
    entrypoint: ["mpi_bootstrap", "role=master", "mpi_master_service_name=master", "mpi_worker_service_name=worker"]
    ports:
      - "${SSH_PORT}:22"
    networks:
      - net
    volumes:
      - /pbs:/pbs

  worker:
    image: $REGISTRY_ADDR:$REGISTRY_PORT/$IMAGE_NAME
    user: root
    entrypoint: ["mpi_bootstrap", "role=worker", "mpi_master_service_name=master", "mpi_worker_service_name=worker"]
    networks:
      - net
    volumes_from:
      - master

networks:
  net:

---

3.将pbsPro安装文件放置在集群的共享目录/pbs下,便于后去安装，共享文件地址为：pbs安装文件

1. 在shell中结合docker-compose编排pbs集群

$# chmod a+x cluster.sh
$# ./cluster.sh up size=3
$# docker ps -a

---

我们发现容器已经自动编排完成
4. 安装openpbs
法一：参照上一节的手动安装办法：利用Dockerfile编排pbspro集群

法二：我们同过放置在共享目录下/pbs下下的安装监本来自动化安装这下文件和过程。

$  cd /pbs/master/
#安装master节点的文件
$# ./login_master.sh cluster_master_1
在docker容器上执行安装pbs文件的shell
$# docker exec -it cluster_master_1 /etc/init.d/pbs status

$# 安装worke节点

$# cd /pbs/worker$# /login_Singleworker cluster_worker_1
$# docker exec -it cluster_master_1 /etc/init.d/pbs status

1. 开启并提交作业
   每个节点开启pbs操作

$# /etc/init.d/pbs start

在maseter上加入slaver1,和slaver2

$# . /etc/profile.d/pbs.sh
$#  qmgr -c 'create node  slaver1'
$#  qmgr -c 'create node  slaver2'

为三个节点添加同意的UID的用户user1每个节点执行如下操作
$# useradd user1 查看$# pbsnodes -a

运行作业：

$# echo 'sleep 10' | qsub
$# qstat -a

至此，整个pbs Pro搭建完毕，此过程可动态调整参数实现集群的扩展。