• 全网最新~使用Shell脚本搭建双机高可用的Lustre集群

    1.环境说明：

    主机名系统挂载情况网卡ens33，Lnet的IP地址Lustre集群名内存
    mds005Centos7.9（共享磁盘）1个mgs，1个MDT，2个OST192.168.10.25/209.21global2G
    mds006Centos7.9（共享磁盘）1个mgs，1个MDT，2个OST192.168.10.26/209.22global2G
    managerCentos7.9无192.168.10.5无2G

    注意：自动化脚本lustre_auto.sh在manager节点上，五块4G共享磁盘（/dev/sdb,/dev/sdc,/dev/sdd,/dev/sde,/dev/sdf），两个网卡（管理IP地址，Lnet通信IP地址）

    2.变量定义：

    1. # ens33上的IP地址，即集群管理IP地址
    2. host_address=(192.168.10.25 192.168.10.26)      
    3. # ens38上的IP地址（其他也可以，两个节点网卡名一样就行），即Lnet通信IP地址
    4. lnet_address=(192.168.209.25 192.168.209.26)    
    5. # 集群节点的主机名和域名
    6. host_hostname=(mds005 mds006) 
    7. # ssh连接密码，集群认证密码
    8. host_passwd=110119

    3.安装expect命令：

    1. # 安装expect命令
    2. expect -v &> /dev/null
    3. if [ `echo $?` -ne 0 ];then
    4.     echo "没有expect，安装expect命令"
    5.     yum install -y expect
    6. fi

    4.配置免密登录：

    1. # 配置免密登录
    2. echo "########################## 本地开始配置ssh ##########################"
    3. if [ `test -a ~/.ssh/id_rsa.pub;echo $?` == 0 ];then
    4. 	echo "ssh公钥已创建"
    5. else
    6. 	echo "ssh公钥未创建，开始创建"
    7. /usr/bin/expect << eof
    8. # 设置捕获字符串后，期待回复的超时时间
    9. set timeout 10
    10.  
    11. spawn ssh-keygen -t rsa -b 1024
    12.  
    13. ## 开始进连续捕获
    14. expect	{
    15.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    16.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    17.         ".ssh/id_rsa)"         { send "\n";  exp_continue }
    18.         "Overwrite (y/n)?"     { send "y\n"; exp_continue }
    19.         "no passphrase):"      { send "\n";  exp_continue }
    20.         "passphrase again:"    { send "\n";  exp_continue }
    21. }
    22. eof
    23. fi
    24.  
    25.  
    26.  
    27. # 本地的密钥开始加入被控制主机
    28. for ((j=0;j<2;j++));do
    29. 	echo "########################## ${host_address[j]}正在被添加公钥 ##########################"
    30. /usr/bin/expect << eof
    31. # 设置捕获字符串后，期待回复的超时时间
    32. set timeout 10
    33.  
    34. spawn ssh-copy-id -i /root/.ssh/id_rsa.pub root@${host_address[j]}
    35.  
    36. ## 开始进连续捕获
    37. expect	{
    38.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    39.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    40. }
    41. eof
    42. 	echo "############# ${host_address[j]}配置完毕 #############"
    43. done
    44.  
    45.  
    46.  
    47.  
    48. # 被控制主机开始创建密钥
    49. for ((j=0;j<2;j++));do
    50. 	echo "########################## ${host_address[j]}开始创建密钥 ##########################"	
    51. 	if [ `ssh root@${host_address[j]} 'test -a ~/.ssh/id_rsa.pub;echo $?'` == 0 ];then
    52. 		echo "ssh公钥已创建"
    53. 	else
    54. 		echo "ssh公钥未创建，开始创建"
    55. /usr/bin/expect << eof
    56. # 设置捕获字符串后，期待回复的超时时间
    57. set timeout 10
    58.  
    59. spawn ssh root@${host_address[j]} "ssh-keygen -t rsa -b 1024"
    60.  
    61. ## 开始进连续捕获
    62. expect	{
    63.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    64.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    65.         ".ssh/id_rsa)"         { send "\n";  exp_continue }
    66.         "Overwrite (y/n)?"     { send "y\n"; exp_continue }
    67.         "no passphrase):"      { send "\n";  exp_continue }
    68.         "passphrase again:"    { send "\n";  exp_continue }
    69. }
    70. eof
    71. 	fi
    72. 	echo "############# ${host_address[j]}配置完毕 #############"
    73. done
    74.  
    75.  
    76.  
    77. # 被控制主机开始分配密钥
    78. for ((j=0;j<2;j++));do
    79. 	for((k=0;k<2;k++));do
    80. 		echo "########################## ${host_address[j]}开始分配公钥给${host_address[k]} ##########################"
    81. /usr/bin/expect << eof
    82. # 设置捕获字符串后，期待回复的超时时间
    83. set timeout 10
    84.  
    85. spawn ssh -t root@${host_address[j]} "ssh root@${host_address[k]}"
    86. spawn ssh -t root@${host_address[j]} "ssh-copy-id -i /root/.ssh/id_rsa.pub root@${host_address[k]}"
    87.  
    88. ## 开始进连续捕获
    89. expect	{
    90.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    91.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    92. }
    93. eof
    94. 		echo "############# ${host_address[j]}配置完毕 #############"
    95. 	done
    96. done

    5.主机名域名映射：

    1. for ((i=0;i<100;i++));do
    2. 	read -p "修改主机名和配置域名映射？(Y/n): " flag
    3. 	if [ "${flag}" == "Y" ];then
    4. 		sleep 3
    5. 		echo "########################## 开始配置主机名和域名映射 ##########################"
    6. 		for ((j=0;j<2;j++));do
    7. 			if [ `ssh root@${host_address[j]} "hostname"` != "${host_hostname[j]}" ];then
    8. 				ssh root@${host_address[j]} "hostnamectl set-hostname ${host_hostname[j]}"
    9. 			fi
    10. 			ssh root@${host_address[j]} "cat << eof > /etc/hosts
    11. 127.0.0.1   localhost localhost.localdomain localhost4 localhost4.localdomain4
    12. ::1         localhost localhost.localdomain localhost6 localhost6.localdomain6
    13. eof"
    14. 			for ((k=0;k<2;k++));do
    15. 				ssh root@${host_address[j]} "echo '${host_address[k]} ${host_hostname[k]}' >> /etc/hosts"
    16. 			done
    17. 		done
    18. 		echo "############# 配置完毕 #############"
    19. 		break
    20. 	elif [ "${flag}" == "n" ];then
    21. 		echo "############# 已跳过步骤-修改主机名和配置域名映射 #############"
    22. 		break
    23. 	elif [ ${i} -eq 99 ];then
    24. 		echo "############# 已退出 #############"
    25. 		exit
    26. 	else continue;fi
    27. done
    28.  
    29. # 测试被控主机ssh连接
    30. for ((j=0;j<2;j++));do
    31. 	echo "########################## ${host_address[j]}开始测试 ##########################"	
    32. 	for ((k=0;k<2;k++));do
    33. /usr/bin/expect << eof
    34. # 设置捕获字符串后，期待回复的超时时间
    35. set timeout 10
    36.  
    37. spawn ssh -t root@${host_address[j]} "ssh root@${host_address[k]} 'exit'"
    38.  
    39. ## 开始进连续捕获
    40. expect	{
    41.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    42. }
    43. eof
    44. 		if [ `echo $?` != 0 ];then
    45. 			echo "${host_hostname[j]}主机无法免密登录${host_hostname[k]}"
    46. 			exit
    47. 		fi
    48. 	done
    49. 	echo "############# ${host_address[j]}测试完毕 #############"
    50. done

    6.配置防火墙和selinux：

    1. # 配置防火墙和selinux
    2. for ((i=0;i<100;i++));do
    3.     read -p "配置防火墙和selinux？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         echo "########################## 开始配置防火墙和selinux ##########################"
    7.         for ((j=0;j<2;j++));do
    8.             ssh root@${host_address[j]} "systemctl stop firewalld;systemctl disable firewalld"
    9.             ssh root@${host_address[j]} "sed -i 's/SELINUX=.*/SELINUX=disabled/' /etc/selinux/config"
    10.         done
    11.         echo "############# 配置完毕 #############"
    12.         break
    13.     elif [ "${flag}" == "n" ];then
    14.         echo "############# 已跳过步骤-配置防火墙和selinux #############"
    15.         break
    16.     elif [ ${i} -eq 99 ];then
    17.         echo "############# 已退出 #############"
    18.         exit
    19.     else continue;fi
    20. done

    7.配置yum源：

    1. # 配置yum源
    2. for ((i=0;i<100;i++));do
    3.     read -p "配置yum源？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         echo "########################## 开始配置yum源 ##########################"
    7.         for ((j=0;j<2;j++));do
    8.             echo "########################## 配置${host_address[j]}的本地yum源 ##########################"
    9.             ssh root@${host_address[j]} "mkdir /mnt/cdrom &> /dev/null;mount /dev/cdrom /mnt/cdrom"
    10.             if [ -z "`ssh root@${host_address[j]} "grep '^\/dev\/cdrom' /etc/fstab"`" ];then
    11.                 ssh root@${host_address[j]} "cat << eof >> /etc/fstab
    12. /dev/cdrom /mnt/cdrom iso9660 defaults  0  0
    13. eof"
    14.             fi
    15.             ssh root@${host_address[j]} "cat << eof > /etc/yum.repos.d/centos-local.repo
    16. [centos7.9]
    17. name=centos7.9
    18. baseurl=file:///mnt/cdrom
    19. enabled=1
    20. gpgcheck=0
    21. eof"
    22.             echo "############# ${host_address[j]}配置完毕 #############"
    23.     
    24.             echo "########################## 配置${host_address[j]}的扩展源 ##########################"
    25.             ssh root@${host_address[j]} "yum install epel-release -y"
    26.             echo "############# ${host_address[j]}配置完毕 #############"
    27.     
    28.             echo "########################## 配置${host_address[j]}的阿里yum源 ##########################"
    29.             ssh root@${host_address[j]} "yum install -y wget"
    30.             if [ `ssh root@${host_address[j]} 'test -a /etc/yum.repos.d/CentOS-Base.repo;echo $?'` == 0 ];then
    31.                 ssh root@${host_address[j]} "wget -O /etc/yum.repos.d/CentOS-Base.repo http://mirrors.aliyun.com/repo/Centos-7.repo"
    32.             fi
    33.             ssh root@${host_address[j]} "yum clean all && yum repolist"
    34.             echo "############# ${host_address[j]}配置完毕 #############"
    35.     
    36.             if [ `ssh root@${host_address[j]} "echo $?"` != 0 ];then
    37.                 echo "yum源配置有误，退出执行脚本"
    38.                 exit
    39.             fi
    40.         done
    41.         break
    42.     elif [ "${flag}" == "n" ];then
    43.         echo "############# 已跳过步骤-配置yum源 #############"
    44.         break
    45.     elif [ ${i} -eq 99 ];then
    46.         echo "############# 已退出 #############"
    47.         exit
    48.     else continue;fi
    49. done

    8.配置chrony时间服务器：

    1. # 配置chrony时间服务器
    2. for ((i=0;i<100;i++));do
    3.     read -p "配置chrony时间服务器？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         echo "########################## 开始配置chrony ##########################"
    7.         for ((j=0;j<2;j++));do
    8.             if [ `ssh root@${host_address[j]} "systemctl restart chronyd;echo $?"` != 0 ];then
    9.                 echo "${host_address[j]} 安装chrony"
    10.                 ssh root@${host_address[j]} "yum install -y chrony && systemctl restart chronyd"
    11.                 if [ `echo $?` != 0 ];then
    12.                     echo "安装失败，请排错!"
    13.                     exit
    14.                 fi
    15.             fi
    16.             echo "${host_address[j]}配置chrony"
    17.                 ssh root@${host_address[j]} "sed -i '/^server/d' /etc/chrony.conf"
    18.             if [ ${host_address[j]} == ${host_address[0]} ];then
    19.                 ssh root@${host_address[j]} "sed -i '2a\server '"${host_address[0]}"' iburst\' /etc/chrony.conf"
    20.                 ssh root@${host_address[j]} "sed -i 's/#allow 192.168.0.0\/16/allow 192.168.10.0\/16/' /etc/chrony.conf"
    21.                 ssh root@${host_address[j]} "sed -i 's/#local stratum 10/local stratum 10/' /etc/chrony.conf"
    22.                 sleep 2
    23. ​
    24.             else
    25.                 ssh root@${host_address[j]} "sed -i '2a\server '"${host_address[0]}"' iburst\' /etc/chrony.conf"
    26.             fi
    27.             ssh root@${host_address[j]} "systemctl restart chronyd && systemctl enable chronyd &> /dev/null"
    28.             sleep 5
    29.             ssh root@${host_address[j]} "timedatectl set-ntp true && chronyc sources -v | sed -n '/^\^\*/p'"
    30.             if [ -z "`ssh root@${host_address[j]} "chronyc sources -v | sed -n '/^\^\*/p'"`" ];then
    31.                 echo -e "\e[31m此节点${host_address[j]}的chrony配置有误，请手动调试\e[0m"
    32.                 exit
    33.             fi
    34.             echo "############# ${host_address[j]}配置完毕 #############"
    35.         done
    36.         break
    37.     elif [ "${flag}" == "n" ];then
    38.         echo "############# 已跳过步骤-配置chrony时间服务器 #############"
    39.         break
    40.     elif [ ${i} -eq 99 ];then
    41.         echo "############# 已退出 #############"
    42.         exit
    43.     else continue;fi
    44. done

    9.安装e2fsprogs：

    1. # 安装e2fsprogs
    2. for ((i=0;i<100;i++));do
    3.     read -p "安装e2fsprogs？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         echo "########################## 开始安装e2fsprogs ##########################"
    7.         for ((j=0;j<2;j++));do
    8.         
    9.             echo "########################## ${host_address[j]}开始安装e2fsprogs ##########################"
    10.             ssh root@${host_address[j]} "rm -rf ~/e2fsprogs1.44.5 && mkdir ~/e2fsprogs1.44.5"
    11.             ssh root@${host_address[j]} "wget -c -r -nd https://downloads.whamcloud.com/public/e2fsprogs/1.44.5.wc1/el7/RPMS/x86_64/ -P ~/e2fsprogs1.44.5"
    12.             ssh root@${host_address[j]} "rm -rf ~/e2fsprogs1.44.5/index.html* ~/e2fsprogs1.44.5/unknown.gif ~/e2fsprogs1.44.5/*.gif ~/e2fsprogs1.44.5/sha256sum"
    13.             ssh root@${host_address[j]} "rpm -Uvh ~/e2fsprogs1.44.5/* --force"
    14.             ssh root@${host_address[j]} "rpm -qa | grep e2fsprogs"
    15.             if [ `echo $?` != 0 ];then
    16.                 echo "安装失败，请排错!"
    17.                 exit
    18.             fi  
    19.             echo "############# ${host_address[j]}配置完毕 #############"
    20.         done
    21.         break
    22.     elif [ "${flag}" == "n" ];then
    23.         echo "############# 已跳过步骤-安装e2fsprogs #############"
    24.         break
    25.     elif [ ${i} -eq 99 ];then
    26.         echo "############# 已退出 #############"
    27.         exit
    28.     else continue;fi
    29. done

    10.安装lustre软件：

    1. for ((i=0;i<100;i++));do
    2.     read -p "安装Lustre软件？(Y/n): " flag
    3.     if [ "${flag}" == "Y" ];then
    4.         sleep 3
    5.         echo "########################## 开始安装lustre ##########################"
    6.         for ((j=0;j<2;j++));do
    7.             echo "########################## ${host_address[j]}开始安装lustre ##########################"
    8.             ssh root@${host_address[j]} "yum install -y linux-firmware dracut selinux-policy-targeted kexec-tools libyaml perl"
    9.             ssh root@${host_address[j]} "rm -rf ~/lustre2.12.1 && mkdir ~/lustre2.12.1"
    10.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/kernel-3.10.0-957.10.1.el7_lustre.x86_64.rpm -P ~/lustre2.12.1"
    11.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/kmod-lustre-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    12.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/kmod-lustre-osd-ldiskfs-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    13.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/lustre-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    14.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/lustre-osd-ldiskfs-mount-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    15.             ssh root@${host_address[j]} "rpm -Uvh ~/lustre2.12.1/*.rpm  --force"
    16.             ssh root@${host_address[j]} "rpm -qa | grep lustre"
    17.             if [ `echo $?` != 0 ];then
    18.                 echo "安装失败，请排错!"
    19.                 exit
    20.             fi
    21.             echo "############# ${host_address[j]}配置完毕 #############"
    22.         done
    23.         break
    24.     elif [ "${flag}" == "n" ];then
    25.         echo "############# 已跳过步骤-安装lustre #############"
    26.         break
    27.     elif [ ${i} -eq 99 ];then
    28.         echo "############# 已退出 #############"
    29.         exit
    30.     else continue;fi
    31. done

    11.控制重启和检测重启：

    1. for ((i=0;i<100;i++));do
    2.     read -p "是否重启集群主机（只有重启kernel内核才能更换生效）？(Y/n): " flag
    3.     if [ "${flag}" == "Y" ];then
    4.         sleep 3
    5.         for((j=0;j<2;j++));do
    6.             ssh root@${host_address[j]} "reboot" 
    7.             continue
    8.         done
    9.     
    10.         for ((k=0;k<100;k++));do
    11.             if [ ${k} -eq 99 ];then
    12.                 echo "############# 设备连接超时.... #############"
    13.                 exit
    14.             fi
    15.             if [ `ssh root@${host_address[0]} -o ConnectTimeout=5 "exit";echo $?` == 0 -a `ssh root@${host_address[1]} -o ConnectTimeout=5 "exit";echo $?` == 0 ];then
    16.                 echo "############# 设备已重启 #############"
    17.                 break
    18.             else
    19.                 echo "############# 设备正在重启 #############"
    20.             fi
    21.         done
    22.         break
    23.     elif [ "${flag}" == "n" ];then
    24.         echo "############# 已跳过步骤-重启 #############"
    25.         break
    26.     elif [ ${i} -eq 99 ];then
    27.         echo "############# 已退出 #############"
    28.         exit
    29.     else continue;fi
    30. done

    12.加载模块：

    1. # 检查lustre
    2. for((i=0;i<2;i++));do
    3.     echo "########################## ${host_address[i]}加载Lustre模块，查看Lustre版本 ##########################"
    4.     ssh root@${host_address[i]} "modprobe lustre && lsmod | grep lustre"
    5.     ssh root@${host_address[i]} "modinfo lustre"
    6.     echo "############# ${host_address[i]}配置完毕 #############"
    7. done

    13.配置和Lustre网络：

    1. # 配置Lnet网卡的IP地址和Lustre网络
    2. for ((i=0;i<100;i++));do
    3.     read -p "是否继续配置Lnet网卡的IP地址？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         for((j=0;j<2;j++));do
    7.             echo "########################## ${host_address[j]}配置Lnet网卡的IP地址 ##########################"
    8.             read -p "请根据以上输出显示中，输入你要配置的LnNet网卡名称：" network_card
    9.             if [ -z "`ssh root@${host_address[j]} "ip addr | grep -o ${network_card}"`" ];then
    10.                 echo "网卡不存在，请重试"
    11.                 exit
    12.             fi
    13.             ssh root@${host_address[j]} "nmcli connection delete ${network_card} &> /dev/null"
    14.             ssh root@${host_address[j]} "nmcli connection add type ethernet con-name ${network_card} ifname ${network_card} ipv4.method manual ipv4.addresses '${lnet_address[j]}/24' autoconnect yes"
    15.             ssh root@${host_address[j]} "nmcli connection up ${network_card}"
    16.             echo "############# ${host_address[j]}配置完毕 #############"
    17.             echo "########################## ${host_address[j]}配置Lnet网络 ##########################"
    18.             ssh root@${host_address[j]} "echo options lnet networks='tcp(ens33),tcp2(${network_card})' > /etc/modprobe.d/lustre.conf"
    19.             ssh root@${host_address[j]} "lustre_rmmod && modprobe -v lustre"
    20.             echo "############# ${host_address[j]}配置完毕 #############"
    21.             echo "############# ${host_address[j]}查看Lnet网络 #############"
    22.             ssh root@${host_address[j]} "lctl list_nids"
    23.             echo "##########################################################"
    24.         done
    25.         break
    26.     elif [ "${flag}" == "n" ];then
    27.         echo "############# 已跳过步骤-配置Lnet网卡的IP地址 #############"
    28.         break
    29.     elif [ ${i} -eq 99 ];then
    30.         echo "############# 已退出 #############"
    31.         exit
    32.     else continue;fi
    33. done

    14.格式化lustre：

    1. # 格式化lustre
    2. for ((i=0;i<100;i++));do
    3.     read -p "是否继续格式化lustre？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         echo "########################## 开始格式化lustre ##########################"
    7.         ssh root@${host_address[0]} "lsblk"
    8.         ssh root@${host_address[0]} "mkfs.lustre --mgs --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdb"
    9.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --mdt --index=0 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdc"
    10.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --mdt --index=1 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdd"
    11.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --ost --index=0 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sde"
    12.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --ost --index=1 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdf"
    13.         echo "############# 格式化完毕 #############"
    14.         echo "########################## 创建挂载点目录 ##########################"
    15.         for((j=0;j<2;j++));do
    16.             ssh root@${host_address[j]} "mkdir /mnt/mgs &> /dev/null;mkdir /mnt/mdt1 &> /dev/null;mkdir /mnt/mdt2 &> /dev/null;mkdir /mnt/ost1 &> /dev/null;mkdir /mnt/ost2 &> /dev/null"
    17.         done
    18.         echo "############# 创建完毕 #############"
    19.         echo "############# ${host_address[1]}查看格式化 #############"
    20.         ssh root@${host_address[1]} "blkid /dev/sdb /dev/sdc /dev/sdd /dev/sde /dev/sdf"
    21.         echo "##########################################################"
    22.         break
    23.     elif [ "${flag}" == "n" ];then
    24.         echo "############# 已跳过步骤-格式化lustre #############"
    25.         break
    26.     elif [ ${i} -eq 99 ];then
    27.         echo "############# 已退出 #############"
    28.         exit
    29.     else continue;fi
    30. done

    15.测试lustre挂载：

    1. # 测试挂载
    2. for ((i=0;i<100;i++));do
    3.     read -p "是否继续测试lustre的挂载？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         for((j=0;j<2;j++));do
    7.         echo "########################## ${host_address[j]}测试挂载 ##########################"
    8.             echo "########################## mgs测试挂载 ##########################"
    9.             ssh root@${host_address[j]} "umount /mnt/mgs &> /dev/null"
    10.             ssh root@${host_address[j]} "mount -t lustre /dev/sdb /mnt/mgs"
    11.             if [ `echo $?` != 0 ];then
    12.                 echo "############# mgs测试有误，请手动查看！ #############"
    13.                 exit
    14.             fi
    15.             echo "########################## mdt1测试挂载 ##########################"
    16.             ssh root@${host_address[j]} "umount /mnt/mdt1 &> /dev/null"
    17.             ssh root@${host_address[j]} "mount -t lustre /dev/sdc /mnt/mdt1"
    18.             if [ `echo $?` != 0 ];then
    19.                 echo "############# mdt1测试有误，请手动查看！ #############"
    20.                 ssh root@${host_address[j]} "umount /mnt/mgs"
    21.                 exit
    22.             fi
    23.             echo "########################## mdt2测试挂载 ##########################"
    24.             ssh root@${host_address[j]} "umount /mnt/mdt2 &> /dev/null"
    25.             ssh root@${host_address[j]} "mount -t lustre /dev/sdd /mnt/mdt2"
    26.             if [ `echo $?` != 0 ];then
    27.                 echo "############# mdt2测试有误，请手动查看！ #############"
    28.                 ssh root@${host_address[j]} "umount /mnt/mdt1;umount /mnt/mgs"
    29.                 exit
    30.             fi
    31.             echo "########################## ost1测试挂载 ##########################"
    32.             ssh root@${host_address[j]} "umount /mnt/ost1 &> /dev/null"
    33.             ssh root@${host_address[j]} "mount -t lustre /dev/sde /mnt/ost1"
    34.             if [ `echo $?` != 0 ];then
    35.                 echo "############# ost1测试有误，请手动查看！ #############"
    36.                 ssh root@${host_address[j]} "umount /mnt/mdt2;umount /mnt/mdt1;umount /mnt/mgs"
    37.                 exit
    38.             fi
    39.             echo "########################## ost2测试挂载 ##########################"
    40.             ssh root@${host_address[j]} "umount /mnt/ost2 &> /dev/null"
    41.             ssh root@${host_address[j]} "mount -t lustre /dev/sdf /mnt/ost2"
    42.             if [ `echo $?` != 0 ];then
    43.                 echo "############# ost2测试有误，请手动查看！ #############"
    44.                 ssh root@${host_address[j]} "umount /mnt/ost1;umount /mnt/mdt2;umount /mnt/mdt1;umount /mnt/mgs"
    45.                 exit
    46.             fi
    47.             ssh root@${host_address[j]} "umount /mnt/ost2;umount /mnt/ost1;umount /mnt/mdt2;umount /mnt/mdt1;umount /mnt/mgs"
    48.         done
    49.         echo "############# 测试完毕 #############"
    50.         break
    51.     elif [ "${flag}" == "n" ];then
    52.         echo "############# 已跳过步骤-测试挂载 #############"
    53.         break
    54.     elif [ ${i} -eq 99 ];then
    55.         echo "############# 已退出 #############"
    56.         exit
    57.     else continue;fi
    58. done

    16.安装高可用和创建集群：

    1. ​
    2. # 安装packemaker和corosync软件和创建集群
    3. for ((i=0;i<100;i++));do
    4.     read -p "是否继续安装packemaker和corosync软件和创建集群？(Y/n): " flag
    5.     if [ "${flag}" == "Y" ];then
    6.         sleep 3
    7.         if [ -z "`ssh root@${host_address[0]} 'pcs status' | grep mycluster`" ];then
    8.             for((j=0;j<2;j++));do
    9.                 echo "########################## ${host_address[j]}开始安装 ##########################"
    10.                 ssh root@${host_address[j]} "yum install pacemaker pcs policycoreutils-python -y"
    11.                 echo "############# ${host_address[j]}安装完毕 #############"
    12.                 echo "########################## ${host_address[j]}开始配置 ##########################"
    13.                 ssh root@${host_address[j]} "systemctl enable pcsd;systemctl restart pcsd"
    14.                 ssh root@${host_address[j]} "echo '${host_passwd}' |passwd --stdin hacluster"
    15.                 echo "############# ${host_address[j]}配置完毕 #############"
    16.             done
    17. /usr/bin/expect << eof
    18. # 设置捕获字符串后，期待回复的超时时间
    19. set timeout 10
    20. ​
    21. spawn ssh root@${host_address[0]} "pcs cluster auth ${host_hostname[*]}"
    22. ​
    23. ## 开始进连续捕获
    24. expect  {
    25.         "Username:"          { send "hacluster\n";  exp_continue }
    26.         "Password:"          { send "${host_passwd}\n"; exp_continue }
    27. }
    28. eof
    29.             echo "########################## 开始创建集群 ##########################"
    30.             ssh root@${host_address[0]} "pcs cluster setup --name mylustre ${host_hostname[*]}"
    31.             echo "############# 创建完毕 #############"
    32.         fi
    33.         echo "########################## 启动集群 ##########################"
    34.         ssh root@${host_address[0]} "pcs cluster start --all"
    35.         echo "############# 启动完毕 #############"
    36.         break
    37.     elif [ "${flag}" == "n" ];then
    38.         echo "############# 已跳过步骤-安装packemaker和corosync软件和配置 #############"
    39.         break
    40.     elif [ ${i} -eq 99 ];then
    41.         echo "############# 已退出 #############"
    42.         exit
    43.     else continue;fi
    44. done
    45.  
    46. ​

    17.配置资源防护:

    1. # 配置资源防护
    2. for ((i=0;i<100;i++));do
    3.     read -p "是否继续配置资源防护？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         echo "########################## 开始配置资源防护 ##########################"
    7.         for((j=0;j<2;j++));do
    8.             ssh root@${host_address[j]} "yum install -y fence-agents-all"
    9.         done
    10.         ssh root@${host_address[0]} "pcs property set stonith-enabled=true"
    11.         if [ `ssh root@${host_address[j]} "pcs status" | grep "stonith:fence_heuristics_ping" | grep -c "Started"` -eq 2 ];then
    12.             echo "############# stonith已创建，并且正常运行，跳过配置stonith #############"
    13.             break
    14.         fi
    15.         if [ `ssh root@${host_address[j]} "pcs status" | grep "stonith:fence_heuristics_ping"` -eq 2 ];then
    16.             ssh root@${host_address[0]} "pcs stonith delete stonith-ping-${host_hostname[0]}"
    17.             ssh root@${host_address[0]} "pcs stonith delete stonith-ping-${host_hostname[0]}"
    18.         fi
    19. ​
    20.         ssh root@${host_address[0]} "pcs stonith create stonith-ping-${host_hostname[0]} fence_heuristics_ping ping_targets=${host_address[0]}"
    21.         ssh root@${host_address[0]} "pcs stonith create stonith-ping-${host_hostname[1]} fence_heuristics_ping ping_targets=${host_address[1]}"
    22.         ssh root@${host_address[0]} "pcs status"
    23.         echo "############# 配置完毕 #############"
    24.         break
    25.     elif [ "${flag}" == "n" ];then
    26.         echo "############# 已跳过步骤-配置资源防护 #############"
    27.         break
    28.     elif [ ${i} -eq 99 ];then
    29.         echo "############# 已退出 #############"
    30.         exit
    31.     else continue;fi
    32. done

    18.创建lustre资源:

    1. # 创建lustre资源
    2. for ((i=0;i<100;i++));do
    3.     read -p "是否继续创建lustre资源（可删除后再创建）？(Y/n): " flag
    4.     if [ "${flag}" == "Y" ];then
    5.         sleep 3
    6.         for((j=0;j<2;j++));do
    7.             if [ -z `ssh root@${host_address[j]} 'rpm -qa | grep lustre-resource-agents-2.12.1-1'` ];then
    8.                 echo "########################## ${host_address[j]}安装ocf:lustre:Lustre包 ##########################"
    9.                 ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/lustre-resource-agents-2.12.1-1.el7.x86_64.rpm"
    10.                 ssh root@${host_address[j]} "rpm -ivh lustre-resource-agents-2.12.1-1.el7.x86_64.rpm"
    11.                 echo "############# 安装完毕 #############"
    12.             fi
    13.         done
    14.         echo "########################## 开始删除lustre资源 ##########################"
    15.         ssh root@${host_address[0]} "pcs resource delete global-ost2 &> /dev/null;pcs resource delete global-ost1 &> /dev/null"
    16.         ssh root@${host_address[0]} "pcs resource delete global-mdt1 &> /dev/null;pcs resource delete global-mdt2 &> /dev/null"
    17.         ssh root@${host_address[0]} "pcs resource delete global-mgs &> /dev/null"
    18.         echo "############# 删除完毕 #############"
    19.         echo "########################## 开始创建lustre资源 ##########################"
    20.         ssh root@${host_address[0]} "pcs resource create global-mgs ocf:lustre:Lustre target=/dev/sdb mountpoint=/mnt/mgs"
    21.         if [ `echo $?` != 0 ];then
    22.             echo "############# mgs资源创建有误，请手动查看！ #############"
    23.             exit
    24.         fi
    25.         ssh root@${host_address[0]} "pcs resource create global-mdt1 ocf:lustre:Lustre target=/dev/sdc mountpoint=/mnt/mdt1"
    26.         if [ `echo $?` != 0 ];then
    27.             echo "############# mdt1资源创建有误，请手动查看！ #############"
    28.             exit
    29.         fi      
    30.         ssh root@${host_address[0]} "pcs resource create global-mdt2 ocf:lustre:Lustre target=/dev/sdd mountpoint=/mnt/mdt2"
    31.         if [ `echo $?` != 0 ];then
    32.             echo "############# mdt2资源创建有误，请手动查看！ #############"
    33.             exit
    34.         fi      
    35.         ssh root@${host_address[0]} "pcs resource create global-ost1 ocf:lustre:Lustre target=/dev/sde mountpoint=/mnt/ost1"
    36.         if [ `echo $?` != 0 ];then
    37.             echo "############# ost1资源创建有误，请手动查看！ #############"
    38.             exit
    39.         fi      
    40.         ssh root@${host_address[0]} "pcs resource create global-ost2 ocf:lustre:Lustre target=/dev/sdf mountpoint=/mnt/ost2"
    41.         if [ `echo $?` != 0 ];then
    42.             echo "############# ost2资源创建有误，请手动查看！ #############"
    43.             exit
    44.         fi      
    45.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-mgs global-mgs ${host_hostname[0]} 10"
    46.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-mdt1 global-mdt1 ${host_hostname[0]} 10"
    47.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-mdt2 global-mdt2 ${host_hostname[1]} 10"
    48.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-ost1 global-ost1 ${host_hostname[0]} 10"
    49.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-ost2 global-ost2 ${host_hostname[1]} 10"
    50.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-mdt1"
    51.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-mdt2"
    52.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-ost1"
    53.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-ost2"
    54.         if [ `echo $?` != 0 ];then
    55.             echo "############# 资源约束创建有误，请手动查看！ #############"
    56.             exit
    57.         fi
    58.         lnet_name1=`ssh root@${host_address[0]} "awk '{print $3}' /etc/modprobe.d/lustre.conf" | awk -F ',' '{print $2}' | awk -F '(' '{print $2}' | awk -F ')' '{print $1}'`
    59.         lnet_name2=`ssh root@${host_address[1]} "awk '{print $3}' /etc/modprobe.d/lustre.conf" | awk -F ',' '{print $2}' | awk -F '(' '{print $2}' | awk -F ')' '{print $1}'`
    60.         ssh root@${host_address[0]} "pcs resource delete ping-lnet &>/dev/null;pcs resource delete global-healthLUSTRE &> /dev/null"
    61.         ssh root@${host_address[0]} "pcs resource create ping-lnet ocf:lustre:healthLNET lctl=true multiplier=1001 device=${lnet_name1} host_list='${lnet_address[0]}@tcp2 ${lnet_address[1]}@tcp2' --clone"
    62.         ssh root@${host_address[0]} "pcs resource create global-healthLUSTRE ocf:lustre:healthLUSTRE --clone"
    63.         if [ `echo $?` != 0 ];then
    64.             echo "############# 资源监听创建有误，请手动查看！ #############"
    65.             exit
    66.         fi
    67.         echo "############# 创建完毕 #############"
    68.         echo "##########################    查看集群状态    ##########################"
    69.         echo "########################################################################"
    70.         echo "########################################################################"
    71.         ssh root@${host_address[0]} "pcs status"
    72.         echo "########################################################################"
    73.         echo "########################################################################"
    74.         echo "################################################## 脚本到此全部执行完毕 ##################################################"
    75.         break
    76.     elif [ "${flag}" == "n" ];then
    77.         echo "############# 已跳过步骤-创建lustre资源 #############"
    78.         break
    79.     elif [ ${i} -eq 99 ];then
    80.         echo "############# 已退出 #############"
    81.         exit
    82.     else continue;fi
    83. done

    19.完整脚本：

    1. # 高可用双机lustre集群的的自动化部署脚本，（部署条件：配置了共享磁盘（/dev/sdb,/dev/sdc,/dev/sdd,/dev/sde,/dev/sdf）。并且有两个网卡，其中一个网卡已经配置了IP地址）
    2. host_address=(192.168.10.25 192.168.10.26)
    3. lnet_address=(192.168.209.25 192.168.209.26)
    4. host_hostname=(mds005 mds006)
    5. host_passwd=110119
    6. ​
    7. ​
    8. # 安装expect命令
    9. expect -v &> /dev/null
    10. if [ `echo $?` -ne 0 ];then
    11.     echo "没有expect，安装expect命令"
    12.     yum install -y expect
    13. fi
    14. ​
    15. # 配置免密登录
    16. echo "########################## 本地开始配置ssh ##########################"
    17. if [ `test -a ~/.ssh/id_rsa.pub;echo $?` == 0 ];then
    18.     echo "ssh公钥已创建"
    19. else
    20.     echo "ssh公钥未创建，开始创建"
    21. /usr/bin/expect << eof
    22. # 设置捕获字符串后，期待回复的超时时间
    23. set timeout 10
    24. ​
    25. spawn ssh-keygen -t rsa -b 1024
    26. ​
    27. ## 开始进连续捕获
    28. expect  {
    29.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    30.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    31.         ".ssh/id_rsa)"         { send "\n";  exp_continue }
    32.         "Overwrite (y/n)?"     { send "y\n"; exp_continue }
    33.         "no passphrase):"      { send "\n";  exp_continue }
    34.         "passphrase again:"    { send "\n";  exp_continue }
    35. }
    36. eof
    37. fi
    38. ​
    39. ​
    40. ​
    41. # 本地的密钥开始加入被控制主机
    42. for ((j=0;j<2;j++));do
    43.     echo "########################## ${host_address[j]}正在被添加公钥 ##########################"
    44. /usr/bin/expect << eof
    45. # 设置捕获字符串后，期待回复的超时时间
    46. set timeout 10
    47. ​
    48. spawn ssh-copy-id -i /root/.ssh/id_rsa.pub root@${host_address[j]}
    49. ​
    50. ## 开始进连续捕获
    51. expect  {
    52.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    53.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    54. }
    55. eof
    56.     echo "############# ${host_address[j]}配置完毕 #############"
    57. done
    58. ​
    59. ​
    60. ​
    61. # 被控制主机开始创建密钥
    62. for ((j=0;j<2;j++));do
    63.     echo "########################## ${host_address[j]}开始创建密钥 ##########################"   
    64.     if [ `ssh root@${host_address[j]} 'test -a ~/.ssh/id_rsa.pub;echo $?'` == 0 ];then
    65.         echo "ssh公钥已创建"
    66.     else
    67.         echo "ssh公钥未创建，开始创建"
    68. /usr/bin/expect << eof
    69. # 设置捕获字符串后，期待回复的超时时间
    70. set timeout 10
    71. ​
    72. spawn ssh root@${host_address[j]} "ssh-keygen -t rsa -b 1024"
    73. ​
    74. ## 开始进连续捕获
    75. expect  {
    76.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    77.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    78.         ".ssh/id_rsa)"         { send "\n";  exp_continue }
    79.         "Overwrite (y/n)?"     { send "y\n"; exp_continue }
    80.         "no passphrase):"      { send "\n";  exp_continue }
    81.         "passphrase again:"    { send "\n";  exp_continue }
    82. }
    83. eof
    84.     fi
    85.     echo "############# ${host_address[j]}配置完毕 #############"
    86. done
    87. ​
    88. ​
    89. ​
    90. # 被控制主机开始分配密钥
    91. for ((j=0;j<2;j++));do
    92.     echo "########################## ${host_address[j]}开始分配公钥 ##########################"   
    93.     for ((k=0;k<2;k++));do
    94.         
    95. /usr/bin/expect << eof
    96. # 设置捕获字符串后，期待回复的超时时间
    97. set timeout 10
    98. ​
    99. spawn ssh root@${host_address[j]} "ssh-copy-id -i /root/.ssh/id_rsa.pub root@${host_address[k]}"
    100. ​
    101. ## 开始进连续捕获
    102. expect  {
    103.         "connecting (yes/no)?" { send "yes\n";  exp_continue }
    104.         "s password:"          { send "${host_passwd}\n"; exp_continue }
    105. }
    106. eof
    107.     done
    108.     echo "############# ${host_address[j]}分配完毕 #############"
    109. done
    110. ​
    111. ​
    112. ​
    113. ​
    114. ​
    115. # 修改主机名和配置域名映射
    116. for ((i=0;i<100;i++));do
    117.     read -p "修改主机名和配置域名映射？(Y/n): " flag
    118.     if [ "${flag}" == "Y" ];then
    119.         sleep 3
    120.         echo "########################## 开始配置主机名和域名映射 ##########################"
    121.         for ((j=0;j<2;j++));do
    122.             if [ `ssh root@${host_address[j]} "hostname"` != "${host_hostname[j]}" ];then
    123.                 ssh root@${host_address[j]} "hostnamectl set-hostname ${host_hostname[j]}"
    124.             fi
    125.             ssh root@${host_address[j]} "cat << eof > /etc/hosts
    126. 127.0.0.1   localhost localhost.localdomain localhost4 localhost4.localdomain4
    127. ::1         localhost localhost.localdomain localhost6 localhost6.localdomain6
    128. eof"
    129.             for ((k=0;k<2;k++));do
    130.                 ssh root@${host_address[j]} "echo '${host_address[k]} ${host_hostname[k]}' >> /etc/hosts"
    131.             done
    132.         done
    133.         echo "############# 配置完毕 #############"
    134.         break
    135.     elif [ "${flag}" == "n" ];then
    136.         echo "############# 已跳过步骤-修改主机名和配置域名映射 #############"
    137.         break
    138.     elif [ ${i} -eq 99 ];then
    139.         echo "############# 已退出 #############"
    140.         exit
    141.     else continue;fi
    142. done
    143. ​
    144. ​
    145. ​
    146. ​
    147. ​
    148. ​
    149. # 配置防火墙和selinux
    150. for ((i=0;i<100;i++));do
    151.     read -p "配置防火墙和selinux？(Y/n): " flag
    152.     if [ "${flag}" == "Y" ];then
    153.         sleep 3
    154.         echo "########################## 开始配置防火墙和selinux ##########################"
    155.         for ((j=0;j<2;j++));do
    156.             ssh root@${host_address[j]} "systemctl stop firewalld;systemctl disable firewalld"
    157.             ssh root@${host_address[j]} "sed -i 's/SELINUX=.*/SELINUX=disabled/' /etc/selinux/config"
    158.         done
    159.         echo "############# 配置完毕 #############"
    160.         break
    161.     elif [ "${flag}" == "n" ];then
    162.         echo "############# 已跳过步骤-配置防火墙和selinux #############"
    163.         break
    164.     elif [ ${i} -eq 99 ];then
    165.         echo "############# 已退出 #############"
    166.         exit
    167.     else continue;fi
    168. done
    169. ​
    170. ​
    171. ​
    172. # 配置yum源
    173. for ((i=0;i<100;i++));do
    174.     read -p "配置yum源？(Y/n): " flag
    175.     if [ "${flag}" == "Y" ];then
    176.         sleep 3
    177.         echo "########################## 开始配置ssh ##########################"
    178.         for ((j=0;j<2;j++));do
    179.             echo "########################## 配置${host_address[j]}的本地yum源 ##########################"
    180.             ssh root@${host_address[j]} "mkdir /mnt/cdrom &> /dev/null;mount /dev/cdrom /mnt/cdrom"
    181.             if [ -z "`ssh root@${host_address[j]} "grep '^\/dev\/cdrom' /etc/fstab"`" ];then
    182.                 ssh root@${host_address[j]} "cat << eof >> /etc/fstab
    183. /dev/cdrom /mnt/cdrom iso9660 defaults  0  0
    184. eof"
    185.             fi
    186.             ssh root@${host_address[j]} "cat << eof > /etc/yum.repos.d/centos-local.repo
    187. [centos7.9]
    188. name=centos7.9
    189. baseurl=file:///mnt/cdrom
    190. enabled=1
    191. gpgcheck=0
    192. eof"
    193.             echo "############# ${host_address[j]}配置完毕 #############"
    194.     
    195.             echo "########################## 配置${host_address[j]}的扩展源 ##########################"
    196.             ssh root@${host_address[j]} "yum install epel-release -y"
    197.             echo "############# ${host_address[j]}配置完毕 #############"
    198.     
    199.             echo "########################## 配置${host_address[j]}的阿里yum源 ##########################"
    200.             ssh root@${host_address[j]} "yum install -y wget"
    201.             if [ `ssh root@${host_address[j]} 'test -a /etc/yum.repos.d/CentOS-Base.repo;echo $?'` == 0 ];then
    202.                 ssh root@${host_address[j]} "wget -O /etc/yum.repos.d/CentOS-Base.repo http://mirrors.aliyun.com/repo/Centos-7.repo"
    203.             fi
    204.             ssh root@${host_address[j]} "yum clean all && yum repolist"
    205.             echo "############# ${host_address[j]}配置完毕 #############"
    206.     
    207.             if [ `ssh root@${host_address[j]} "echo $?"` != 0 ];then
    208.                 echo "yum源配置有误，退出执行脚本"
    209.                 exit
    210.             fi
    211.         done
    212.         break
    213.     elif [ "${flag}" == "n" ];then
    214.         echo "############# 已跳过步骤-配置yum源 #############"
    215.         break
    216.     elif [ ${i} -eq 99 ];then
    217.         echo "############# 已退出 #############"
    218.         exit
    219.     else continue;fi
    220. done
    221. ​
    222. ​
    223. # 配置chrony时间服务器
    224. for ((i=0;i<100;i++));do
    225.     read -p "配置chrony时间服务器？(Y/n): " flag
    226.     if [ "${flag}" == "Y" ];then
    227.         sleep 3
    228.         echo "########################## 开始配置chrony ##########################"
    229.         for ((j=0;j<2;j++));do
    230.             if [ `ssh root@${host_address[j]} "systemctl restart chronyd;echo $?"` != 0 ];then
    231.                 echo "${host_address[j]} 安装chrony"
    232.                 ssh root@${host_address[j]} "yum install -y chrony && systemctl restart chronyd"
    233.                 if [ `echo $?` != 0 ];then
    234.                     echo "安装失败，请排错!"
    235.                     exit
    236.                 fi
    237.             fi
    238.             echo "${host_address[j]}配置chrony"
    239.                 ssh root@${host_address[j]} "sed -i '/^server/d' /etc/chrony.conf"
    240.             if [ ${host_address[j]} == ${host_address[0]} ];then
    241.                 ssh root@${host_address[j]} "sed -i '2a\server '"${host_address[0]}"' iburst\' /etc/chrony.conf"
    242.                 ssh root@${host_address[j]} "sed -i 's/#allow 192.168.0.0\/16/allow 192.168.10.0\/16/' /etc/chrony.conf"
    243.                 ssh root@${host_address[j]} "sed -i 's/#local stratum 10/local stratum 10/' /etc/chrony.conf"
    244.                 sleep 2
    245. ​
    246.             else
    247.                 ssh root@${host_address[j]} "sed -i '2a\server '"${host_address[0]}"' iburst\' /etc/chrony.conf"
    248.             fi
    249.             ssh root@${host_address[j]} "systemctl restart chronyd && systemctl enable chronyd &> /dev/null"
    250.             sleep 5
    251.             ssh root@${host_address[j]} "timedatectl set-ntp true && chronyc sources -v | sed -n '/^\^\*/p'"
    252.             if [ -z "`ssh root@${host_address[j]} "chronyc sources -v | sed -n '/^\^\*/p'"`" ];then
    253.                 echo -e "\e[31m此节点${host_address[j]}的chrony配置有误，请手动调试\e[0m"
    254.                 exit
    255.             fi
    256.             echo "############# ${host_address[j]}配置完毕 #############"
    257.         done
    258.         break
    259.     elif [ "${flag}" == "n" ];then
    260.         echo "############# 已跳过步骤-配置chrony时间服务器 #############"
    261.         break
    262.     elif [ ${i} -eq 99 ];then
    263.         echo "############# 已退出 #############"
    264.         exit
    265.     else continue;fi
    266. done
    267. ​
    268. ​
    269. ​
    270. ​
    271. ​
    272. ​
    273. # 安装e2fsprogs
    274. for ((i=0;i<100;i++));do
    275.     read -p "安装e2fsprogs？(Y/n): " flag
    276.     if [ "${flag}" == "Y" ];then
    277.         sleep 3
    278.         echo "########################## 开始安装e2fsprogs ##########################"
    279.         for ((j=0;j<2;j++));do
    280.         
    281.             echo "########################## ${host_address[j]}开始安装e2fsprogs ##########################"
    282.             ssh root@${host_address[j]} "rm -rf ~/e2fsprogs1.44.5 && mkdir ~/e2fsprogs1.44.5"
    283.             ssh root@${host_address[j]} "wget -c -r -nd https://downloads.whamcloud.com/public/e2fsprogs/1.44.5.wc1/el7/RPMS/x86_64/ -P ~/e2fsprogs1.44.5"
    284.             ssh root@${host_address[j]} "rm -rf ~/e2fsprogs1.44.5/index.html* ~/e2fsprogs1.44.5/unknown.gif ~/e2fsprogs1.44.5/*.gif ~/e2fsprogs1.44.5/sha256sum"
    285.             ssh root@${host_address[j]} "rpm -Uvh ~/e2fsprogs1.44.5/* --force"
    286.             ssh root@${host_address[j]} "rpm -qa | grep e2fsprogs"
    287.             if [ `echo $?` != 0 ];then
    288.                 echo "安装失败，请排错!"
    289.                 exit
    290.             fi  
    291.             echo "############# ${host_address[j]}配置完毕 #############"
    292.         done
    293.         break
    294.     elif [ "${flag}" == "n" ];then
    295.         echo "############# 已跳过步骤-安装e2fsprogs #############"
    296.         break
    297.     elif [ ${i} -eq 99 ];then
    298.         echo "############# 已退出 #############"
    299.         exit
    300.     else continue;fi
    301. done
    302. ​
    303. ​
    304. ​
    305. ​
    306. # 安装lustre
    307. for ((i=0;i<100;i++));do
    308.     read -p "安装Lustre软件？(Y/n): " flag
    309.     if [ "${flag}" == "Y" ];then
    310.         sleep 3
    311.         echo "########################## 开始安装lustre ##########################"
    312.         for ((j=0;j<2;j++));do
    313.             echo "########################## ${host_address[j]}开始安装lustre ##########################"
    314.             ssh root@${host_address[j]} "yum install -y linux-firmware dracut selinux-policy-targeted kexec-tools libyaml perl"
    315.             ssh root@${host_address[j]} "rm -rf ~/lustre2.12.1 && mkdir ~/lustre2.12.1"
    316.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/kernel-3.10.0-957.10.1.el7_lustre.x86_64.rpm -P ~/lustre2.12.1"
    317.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/kmod-lustre-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    318.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/kmod-lustre-osd-ldiskfs-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    319.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/lustre-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    320.             ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/lustre-osd-ldiskfs-mount-2.12.1-1.el7.x86_64.rpm -P ~/lustre2.12.1"
    321.             ssh root@${host_address[j]} "rpm -Uvh ~/lustre2.12.1/*.rpm  --force"
    322.             ssh root@${host_address[j]} "rpm -qa | grep lustre"
    323.             if [ `echo $?` != 0 ];then
    324.                 echo "安装失败，请排错!"
    325.                 exit
    326.             fi
    327.             echo "############# ${host_address[j]}配置完毕 #############"
    328.         done
    329.         break
    330.     elif [ "${flag}" == "n" ];then
    331.         echo "############# 已跳过步骤-安装lustre #############"
    332.         break
    333.     elif [ ${i} -eq 99 ];then
    334.         echo "############# 已退出 #############"
    335.         exit
    336.     else continue;fi
    337. done
    338. ​
    339. ​
    340. ​
    341. #echo "########################## 请手动重启 ##########################"
    342. for ((i=0;i<100;i++));do
    343.     read -p "是否重启集群主机（只有重启kernel内核才能更换生效）？(Y/n): " flag
    344.     if [ "${flag}" == "Y" ];then
    345.         sleep 3
    346.         for((j=0;j<2;j++));do
    347.             ssh root@${host_address[j]} "reboot" 
    348.             continue
    349.         done
    350.     
    351.         for ((k=0;k<100;k++));do
    352.             if [ ${k} -eq 99 ];then
    353.                 echo "############# 设备连接超时.... #############"
    354.                 exit
    355.             fi
    356.             if [ `ssh root@${host_address[0]} -o ConnectTimeout=5 "exit";echo $?` == 0 -a `ssh root@${host_address[1]} -o ConnectTimeout=5 "exit";echo $?` == 0 ];then
    357.                 echo "############# 设备已重启 #############"
    358.                 break
    359.             else
    360.                 echo "############# 设备正在重启 #############"
    361.             fi
    362.         done
    363.         break
    364.     elif [ "${flag}" == "n" ];then
    365.         echo "############# 已跳过步骤-重启 #############"
    366.         break
    367.     elif [ ${i} -eq 99 ];then
    368.         echo "############# 已退出 #############"
    369.         exit
    370.     else continue;fi
    371. done
    372. ​
    373. sleep 5
    374. ​
    375. ​
    376. # 检查lustre
    377. for((i=0;i<2;i++));do
    378.     echo "########################## ${host_address[i]}加载Lustre模块，查看Lustre版本 ##########################"
    379.     ssh root@${host_address[i]} "modprobe lustre && lsmod | grep lustre"
    380.     ssh root@${host_address[i]} "modinfo lustre"
    381.     echo "############# ${host_address[i]}配置完毕 #############"
    382. done
    383. ​
    384. ​
    385. ​
    386. # 配置配置Lnet网卡的IP地址和Lustre网络
    387. for ((i=0;i<100;i++));do
    388.     read -p "是否继续配置Lnet网卡的IP地址？(Y/n): " flag
    389.     if [ "${flag}" == "Y" ];then
    390.         sleep 3
    391.         for((j=0;j<2;j++));do
    392.             echo "########################## ${host_address[j]}配置Lnet网卡的IP地址 ##########################"
    393.             read -p "请根据以上输出显示中，输入你要配置的LnNet网卡名称：" network_card
    394.             if [ -z "`ssh root@${host_address[j]} "ip addr | grep -o ${network_card}"`" ];then
    395.                 echo "网卡不存在，请重试"
    396.                 exit
    397.             fi
    398.             ssh root@${host_address[j]} "nmcli connection delete ${network_card} &> /dev/null"
    399.             ssh root@${host_address[j]} "nmcli connection add type ethernet con-name ${network_card} ifname ${network_card} ipv4.method manual ipv4.addresses '${lnet_address[j]}/24' autoconnect yes"
    400.             ssh root@${host_address[j]} "nmcli connection up ${network_card}"
    401.             echo "############# ${host_address[j]}配置完毕 #############"
    402.             echo "########################## ${host_address[j]}配置Lnet网络 ##########################"
    403.             ssh root@${host_address[j]} "echo options lnet networks='tcp(ens33),tcp2(${network_card})' > /etc/modprobe.d/lustre.conf"
    404.             ssh root@${host_address[j]} "lustre_rmmod && modprobe -v lustre"
    405.             echo "############# ${host_address[j]}配置完毕 #############"
    406.             echo "############# ${host_address[j]}查看Lnet网络 #############"
    407.             ssh root@${host_address[j]} "lctl list_nids"
    408.             echo "##########################################################"
    409.         done
    410.         break
    411.     elif [ "${flag}" == "n" ];then
    412.         echo "############# 已跳过步骤-配置Lnet网卡的IP地址 #############"
    413.         break
    414.     elif [ ${i} -eq 99 ];then
    415.         echo "############# 已退出 #############"
    416.         exit
    417.     else continue;fi
    418. done
    419. ​
    420. ​
    421. ​
    422. ​
    423. ​
    424. # 格式化lustre
    425. for ((i=0;i<100;i++));do
    426.     read -p "是否继续格式化lustre？(Y/n): " flag
    427.     if [ "${flag}" == "Y" ];then
    428.         sleep 3
    429.         echo "########################## 开始格式化lustre ##########################"
    430.         ssh root@${host_address[0]} "lsblk"
    431.         ssh root@${host_address[0]} "mkfs.lustre --mgs --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdb"
    432.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --mdt --index=0 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdc"
    433.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --mdt --index=1 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdd"
    434.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --ost --index=0 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sde"
    435.         ssh root@${host_address[0]} "mkfs.lustre --fsname global --ost --index=1 --servicenode=${lnet_address[0]}@tcp2 --servicenode=${lnet_address[1]}@tcp2 --mgsnode=${lnet_address[0]}@tcp2 --mgsnode=${lnet_address[1]}@tcp2 --backfstype=ldiskfs --reformat /dev/sdf"
    436.         echo "############# 格式化完毕 #############"
    437.         echo "########################## 创建挂载点目录 ##########################"
    438.         for((j=0;j<2;j++));do
    439.             ssh root@${host_address[j]} "mkdir /mnt/mgs &> /dev/null;mkdir /mnt/mdt1 &> /dev/null;mkdir /mnt/mdt2 &> /dev/null;mkdir /mnt/ost1 &> /dev/null;mkdir /mnt/ost2 &> /dev/null"
    440.         done
    441.         echo "############# 创建完毕 #############"
    442.         echo "############# ${host_address[1]}查看格式化 #############"
    443.         ssh root@${host_address[1]} "blkid /dev/sdb /dev/sdc /dev/sdd /dev/sde /dev/sdf"
    444.         echo "##########################################################"
    445.         break
    446.     elif [ "${flag}" == "n" ];then
    447.         echo "############# 已跳过步骤-格式化lustre #############"
    448.         break
    449.     elif [ ${i} -eq 99 ];then
    450.         echo "############# 已退出 #############"
    451.         exit
    452.     else continue;fi
    453. done
    454. ​
    455. ​
    456. ​
    457. # 测试挂载
    458. for ((i=0;i<100;i++));do
    459.     read -p "是否继续测试lustre的挂载？(Y/n): " flag
    460.     if [ "${flag}" == "Y" ];then
    461.         sleep 3
    462.         for((j=0;j<2;j++));do
    463.         echo "########################## ${host_address[j]}测试挂载 ##########################"
    464.             echo "########################## mgs测试挂载 ##########################"
    465.             ssh root@${host_address[j]} "umount /mnt/mgs &> /dev/null"
    466.             ssh root@${host_address[j]} "mount -t lustre /dev/sdb /mnt/mgs"
    467.             if [ `echo $?` != 0 ];then
    468.                 echo "############# mgs测试有误，请手动查看！ #############"
    469.                 exit
    470.             fi
    471.             echo "########################## mdt1测试挂载 ##########################"
    472.             ssh root@${host_address[j]} "umount /mnt/mdt1 &> /dev/null"
    473.             ssh root@${host_address[j]} "mount -t lustre /dev/sdc /mnt/mdt1"
    474.             if [ `echo $?` != 0 ];then
    475.                 echo "############# mdt1测试有误，请手动查看！ #############"
    476.                 ssh root@${host_address[j]} "umount /mnt/mgs"
    477.                 exit
    478.             fi
    479.             echo "########################## mdt2测试挂载 ##########################"
    480.             ssh root@${host_address[j]} "umount /mnt/mdt2 &> /dev/null"
    481.             ssh root@${host_address[j]} "mount -t lustre /dev/sdd /mnt/mdt2"
    482.             if [ `echo $?` != 0 ];then
    483.                 echo "############# mdt2测试有误，请手动查看！ #############"
    484.                 ssh root@${host_address[j]} "umount /mnt/mdt1;umount /mnt/mgs"
    485.                 exit
    486.             fi
    487.             echo "########################## ost1测试挂载 ##########################"
    488.             ssh root@${host_address[j]} "umount /mnt/ost1 &> /dev/null"
    489.             ssh root@${host_address[j]} "mount -t lustre /dev/sde /mnt/ost1"
    490.             if [ `echo $?` != 0 ];then
    491.                 echo "############# ost1测试有误，请手动查看！ #############"
    492.                 ssh root@${host_address[j]} "umount /mnt/mdt2;umount /mnt/mdt1;umount /mnt/mgs"
    493.                 exit
    494.             fi
    495.             echo "########################## ost2测试挂载 ##########################"
    496.             ssh root@${host_address[j]} "umount /mnt/ost2 &> /dev/null"
    497.             ssh root@${host_address[j]} "mount -t lustre /dev/sdf /mnt/ost2"
    498.             if [ `echo $?` != 0 ];then
    499.                 echo "############# ost2测试有误，请手动查看！ #############"
    500.                 ssh root@${host_address[j]} "umount /mnt/ost1;umount /mnt/mdt2;umount /mnt/mdt1;umount /mnt/mgs"
    501.                 exit
    502.             fi
    503.             ssh root@${host_address[j]} "umount /mnt/ost2;umount /mnt/ost1;umount /mnt/mdt2;umount /mnt/mdt1;umount /mnt/mgs"
    504.         done
    505.         echo "############# 测试完毕 #############"
    506.         break
    507.     elif [ "${flag}" == "n" ];then
    508.         echo "############# 已跳过步骤-测试挂载 #############"
    509.         break
    510.     elif [ ${i} -eq 99 ];then
    511.         echo "############# 已退出 #############"
    512.         exit
    513.     else continue;fi
    514. done
    515. ​
    516. ​
    517. ​
    518. ​
    519. # 安装packemaker和corosync软件和创建集群
    520. for ((i=0;i<100;i++));do
    521.     read -p "是否继续安装packemaker和corosync软件和创建集群？(Y/n): " flag
    522.     if [ "${flag}" == "Y" ];then
    523.         sleep 3
    524.         if [ -z "`ssh root@${host_address[0]} 'pcs status' | grep mycluster`" ];then
    525.             for((j=0;j<2;j++));do
    526.                 echo "########################## ${host_address[j]}开始安装 ##########################"
    527.                 ssh root@${host_address[j]} "yum install pacemaker pcs policycoreutils-python -y"
    528.                 echo "############# ${host_address[j]}安装完毕 #############"
    529.                 echo "########################## ${host_address[j]}开始配置 ##########################"
    530.                 ssh root@${host_address[j]} "systemctl enable pcsd;systemctl restart pcsd"
    531.                 ssh root@${host_address[j]} "echo '${host_passwd}' |passwd --stdin hacluster"
    532.                 echo "############# ${host_address[j]}配置完毕 #############"
    533.             done
    534. /usr/bin/expect << eof
    535. # 设置捕获字符串后，期待回复的超时时间
    536. set timeout 10
    537. ​
    538. spawn ssh root@${host_address[0]} "pcs cluster auth ${host_hostname[*]}"
    539. ​
    540. ## 开始进连续捕获
    541. expect  {
    542.         "Username:"          { send "hacluster\n";  exp_continue }
    543.         "Password:"          { send "${host_passwd}\n"; exp_continue }
    544. }
    545. eof
    546.             echo "########################## 开始创建集群 ##########################"
    547.             ssh root@${host_address[0]} "pcs cluster setup --name mylustre ${host_hostname[*]}"
    548.             echo "############# 创建完毕 #############"
    549.         fi
    550.         echo "########################## 启动集群 ##########################"
    551.         ssh root@${host_address[0]} "pcs cluster start --all"
    552.         echo "############# 启动完毕 #############"
    553.         break
    554.     elif [ "${flag}" == "n" ];then
    555.         echo "############# 已跳过步骤-安装packemaker和corosync软件和配置 #############"
    556.         break
    557.     elif [ ${i} -eq 99 ];then
    558.         echo "############# 已退出 #############"
    559.         exit
    560.     else continue;fi
    561. done
    562. ​
    563. ​
    564. ​
    565. # 配置资源防护
    566. for ((i=0;i<100;i++));do
    567.     read -p "是否继续配置资源防护？(Y/n): " flag
    568.     if [ "${flag}" == "Y" ];then
    569.         sleep 3
    570.         echo "########################## 开始配置资源防护 ##########################"
    571.         for((j=0;j<2;j++));do
    572.             ssh root@${host_address[j]} "yum install -y fence-agents-all"
    573.         done
    574.         ssh root@${host_address[0]} "pcs property set stonith-enabled=true"
    575.         if [ `ssh root@${host_address[j]} "pcs status" | grep "stonith:fence_heuristics_ping" | grep -c "Started"` -eq 2 ];then
    576.             echo "############# stonith已创建，并且正常运行，跳过配置stonith #############"
    577.             break
    578.         fi
    579.         if [ `ssh root@${host_address[j]} "pcs status" | grep "stonith:fence_heuristics_ping"` -eq 2 ];then
    580.             ssh root@${host_address[0]} "pcs stonith delete stonith-ping-${host_hostname[0]}"
    581.             ssh root@${host_address[0]} "pcs stonith delete stonith-ping-${host_hostname[0]}"
    582.         fi
    583. ​
    584.         ssh root@${host_address[0]} "pcs stonith create stonith-ping-${host_hostname[0]} fence_heuristics_ping ping_targets=${host_address[0]}"
    585.         ssh root@${host_address[0]} "pcs stonith create stonith-ping-${host_hostname[1]} fence_heuristics_ping ping_targets=${host_address[1]}"
    586.         ssh root@${host_address[0]} "pcs status"
    587.         echo "############# 配置完毕 #############"
    588.         break
    589.     elif [ "${flag}" == "n" ];then
    590.         echo "############# 已跳过步骤-配置资源防护 #############"
    591.         break
    592.     elif [ ${i} -eq 99 ];then
    593.         echo "############# 已退出 #############"
    594.         exit
    595.     else continue;fi
    596. done
    597. ​
    598. ​
    599. # 创建lustre资源
    600. for ((i=0;i<100;i++));do
    601.     read -p "是否继续创建lustre资源（可删除后再创建）？(Y/n): " flag
    602.     if [ "${flag}" == "Y" ];then
    603.         sleep 3
    604.         for((j=0;j<2;j++));do
    605.             if [ -z `ssh root@${host_address[j]} 'rpm -qa | grep lustre-resource-agents-2.12.1-1'` ];then
    606.                 echo "########################## ${host_address[j]}安装ocf:lustre:Lustre包 ##########################"
    607.                 ssh root@${host_address[j]} "wget https://downloads.whamcloud.com/public/lustre/lustre-2.12.1/el7/server/RPMS/x86_64/lustre-resource-agents-2.12.1-1.el7.x86_64.rpm"
    608.                 ssh root@${host_address[j]} "rpm -ivh lustre-resource-agents-2.12.1-1.el7.x86_64.rpm"
    609.                 echo "############# 安装完毕 #############"
    610.             fi
    611.         done
    612.         echo "########################## 开始删除lustre资源 ##########################"
    613.         ssh root@${host_address[0]} "pcs resource delete global-ost2 &> /dev/null;pcs resource delete global-ost1 &> /dev/null"
    614.         ssh root@${host_address[0]} "pcs resource delete global-mdt1 &> /dev/null;pcs resource delete global-mdt2 &> /dev/null"
    615.         ssh root@${host_address[0]} "pcs resource delete global-mgs &> /dev/null"
    616.         echo "############# 删除完毕 #############"
    617.         echo "########################## 开始创建lustre资源 ##########################"
    618.         ssh root@${host_address[0]} "pcs resource create global-mgs ocf:lustre:Lustre target=/dev/sdb mountpoint=/mnt/mgs"
    619.         if [ `echo $?` != 0 ];then
    620.             echo "############# mgs资源创建有误，请手动查看！ #############"
    621.             exit
    622.         fi
    623.         ssh root@${host_address[0]} "pcs resource create global-mdt1 ocf:lustre:Lustre target=/dev/sdc mountpoint=/mnt/mdt1"
    624.         if [ `echo $?` != 0 ];then
    625.             echo "############# mdt1资源创建有误，请手动查看！ #############"
    626.             exit
    627.         fi      
    628.         ssh root@${host_address[0]} "pcs resource create global-mdt2 ocf:lustre:Lustre target=/dev/sdd mountpoint=/mnt/mdt2"
    629.         if [ `echo $?` != 0 ];then
    630.             echo "############# mdt2资源创建有误，请手动查看！ #############"
    631.             exit
    632.         fi      
    633.         ssh root@${host_address[0]} "pcs resource create global-ost1 ocf:lustre:Lustre target=/dev/sde mountpoint=/mnt/ost1"
    634.         if [ `echo $?` != 0 ];then
    635.             echo "############# ost1资源创建有误，请手动查看！ #############"
    636.             exit
    637.         fi      
    638.         ssh root@${host_address[0]} "pcs resource create global-ost2 ocf:lustre:Lustre target=/dev/sdf mountpoint=/mnt/ost2"
    639.         if [ `echo $?` != 0 ];then
    640.             echo "############# ost2资源创建有误，请手动查看！ #############"
    641.             exit
    642.         fi      
    643.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-mgs global-mgs ${host_hostname[0]} 10"
    644.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-mdt1 global-mdt1 ${host_hostname[0]} 10"
    645.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-mdt2 global-mdt2 ${host_hostname[1]} 10"
    646.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-ost1 global-ost1 ${host_hostname[0]} 10"
    647.         ssh root@${host_address[0]} "pcs constraint location add global-constraint-ost2 global-ost2 ${host_hostname[1]} 10"
    648.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-mdt1"
    649.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-mdt2"
    650.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-ost1"
    651.         ssh root@${host_address[0]} "pcs constraint order start global-mgs then start global-ost2"
    652.         if [ `echo $?` != 0 ];then
    653.             echo "############# 资源约束创建有误，请手动查看！ #############"
    654.             exit
    655.         fi
    656.         lnet_name1=`ssh root@${host_address[0]} "awk '{print $3}' /etc/modprobe.d/lustre.conf" | awk -F ',' '{print $2}' | awk -F '(' '{print $2}' | awk -F ')' '{print $1}'`
    657.         lnet_name2=`ssh root@${host_address[1]} "awk '{print $3}' /etc/modprobe.d/lustre.conf" | awk -F ',' '{print $2}' | awk -F '(' '{print $2}' | awk -F ')' '{print $1}'`
    658.         ssh root@${host_address[0]} "pcs resource delete ping-lnet &>/dev/null;pcs resource delete global-healthLUSTRE &> /dev/null"
    659.         ssh root@${host_address[0]} "pcs resource create ping-lnet ocf:lustre:healthLNET lctl=true multiplier=1001 device=${lnet_name1} host_list='${lnet_address[0]}@tcp2 ${lnet_address[1]}@tcp2' --clone"
    660.         ssh root@${host_address[0]} "pcs resource create global-healthLUSTRE ocf:lustre:healthLUSTRE --clone"
    661.         if [ `echo $?` != 0 ];then
    662.             echo "############# 资源监听创建有误，请手动查看！ #############"
    663.             exit
    664.         fi
    665.         echo "############# 创建完毕 #############"
    666.         echo "##########################    查看集群状态    ##########################"
    667.         echo "########################################################################"
    668.         echo "########################################################################"
    669.         ssh root@${host_address[0]} "pcs status"
    670.         echo "########################################################################"
    671.         echo "########################################################################"
    672.         echo "################################################## 脚本到此全部执行完毕 ##################################################"
    673.         break
    674.     elif [ "${flag}" == "n" ];then
    675.         echo "############# 已跳过步骤-创建lustre资源 #############"
    676.         break
    677.     elif [ ${i} -eq 99 ];then
    678.         echo "############# 已退出 #############"
    679.         exit
    680.     else continue;fi
    681. done

  • 相关阅读:
    jmeter 用户自定义变量
    Linux 防火墙 firewalld 常用命令
    [补题记录] Atcoder Beginner Contest 300（E）
    【Flask】官方教程(Tutorial)-part3：blog蓝图、项目可安装化
    国外访问学者申请政策解析
    CSS_文本属性
    golang常用包
    KNN学习代码理解尝试
    【OpenSSL】Ubuntu 下编译OpenSSL
    Qt5开发及实例V2.0-第十五章-Qt单元测试框架
  • 原文地址：https://blog.csdn.net/qq_56776641/article/details/133300332