systemctl disable firewalld.service
systemctl stop firewalld.service

setenforce 0
sed -i 's/^SELINUX=.*$/SELINUX=disabled/'  /etc/selinux/config

# 已使用国内yum镜像，原镜像地址是http://yum.mariadb.org
echo '
[mariadb]
name = MariaDB
baseurl = http://mirrors.ustc.edu.cn/mariadb/yum/10.1/centos7-amd64
gpgkey=http://mirrors.ustc.edu.cn/mariadb/yum/RPM-GPG-KEY-MariaDB
gpgcheck=1' > /etc/yum.repos.d/MariaDB.repo

yum install -y mariadb mariadb-server mariadb-common galera rsync

systemctl start mariadb
mysql_secure_installation # 注意这一步是有交互的，需要回答一些问题，做一些设置
systemctl stop mariadb

sudo -u mysql /usr/sbin/mysqld --wsrep-new-cluster &> /tmp/wsrep_new_cluster.log &
disown $!
tail -f /tmp/wsrep_new_cluster.log

systemctl start mariadb

mysql -e "show status like 'wsrep_cluster_size'"  # 这里应该显示集群里有3个节点
mysql -e "show status like 'wsrep_connected'"     # 这里应该显示ON
mysql -e "show status like 'wsrep_incoming_addresses'" # 这里应该显示10.211.55.7:3306,10.211.55.8:3306,10.211.55.6:3306
mysql -e "show status like 'wsrep_local_state_comment'" # 这里节点的同步状态

mysql -e "show status like 'wsrep_%'"

cat /usr/local/bin/mariadb_cluster_helper.sh

#!/bin/bash
GRASTATE_FILE=/var/lib/mysql/grastate.dat
WSREP_NEW_CLUSTER_LOG_FILE=/tmp/wsrep_new_cluster.log
# 如果启动mariadb超过10秒还没返回0，则认为失败了
START_MARIADB_TIMEOUT=10
# 以--wsrep-new-cluster参数启动，超过5次检查，发现仍没有其它节点加入集群，则认为此路不通
SPECIAL_START_WAIT_MAX_COUNT=5
# 得到本机IP
MY_IP=$(grep 'wsrep_node_address' /etc/my.cnf.d/server.cnf | awk -F '=' '{print $2}')

# 杀掉mysqld进程
function kill_mysqld_process() {
    (ps -ef|grep mysqld|grep -v grep|awk '{print $2}'|xargs kill -9) &>/dev/null
}

# 正常启动mariadb
function start_mariadb_normal(){
    # 首先确保safe_to_bootstrap标记为0
    sed -i 's/^safe_to_bootstrap.*$/safe_to_bootstrap: 0/' $GRASTATE_FILE
    timeout $START_MARIADB_TIMEOUT systemctl start mariadb &> /dev/null
    return $?
}

# 以--wsrep-new-cluster参数启动mariadb
function start_mariadb_special(){
    # 首先确保safe_to_bootstrap标记为1
    sed -i 's/^safe_to_bootstrap.*$/safe_to_bootstrap: 1/' $GRASTATE_FILE
    # 以--wsrep-new-cluster参数启动mariadb
    /usr/sbin/mysqld --user=mysql --wsrep-new-cluster &> $WSREP_NEW_CLUSTER_LOG_FILE &
    disown $!
    try_count=0
    # 循环检查
    while [ 1 ]; do
        # 如果超过SPECIAL_START_WAIT_MAX_COUNT次检查，仍没有其它节点加入集群，则认为此路不通，尝试正常启动，跳出循环
        if [ $try_count -gt $SPECIAL_START_WAIT_MAX_COUNT ] ; then
            kill_mysqld_process
            start_mariadb_normal
            return $?
        fi
        new_joined_count=$(grep 'synced with group' /tmp/wsrep_new_cluster.log | grep -v $MY_IP|wc -l)
        exception_count=$(grep 'exception from gcomm, backend must be restarted' $WSREP_NEW_CLUSTER_LOG_FILE | wc -l)
        # 如果新加入的节点数大于0，则认为集群就绪了，可正常启动了，跳出循环
        # 如果运行日志中发现了异常(两个节点都以--wsrep-new-cluster参数启动，其中一个会报错)，则认为此路不通，尝试正常启动，跳出循环
        if [ $new_joined_count -gt 0 ] || [ $exception_count -gt 0 ] ; then
            kill_mysqld_process
            start_mariadb_normal
            return $?
        else
            try_count=$(( $try_count + 1 ))
        fi
        sleep 5
    done
}

# 首先杀掉mysqld进程
kill_mysqld_process

ret=-1

# 如果safe_to_bootstrap标记为1，则立即以--wsrep-new-cluster参数启动
if [ -f $GRASTATE_FILE ]; then
    safe_bootstrap_flag=$(grep 'safe_to_bootstrap' $GRASTATE_FILE | awk -F ': ' '{print $2}')
    if [ $safe_bootstrap_flag -eq 1 ] ; then
        start_mariadb_special
        ret=$?
    else
        start_mariadb_normal
        ret=$?
    fi
else
    start_mariadb_normal
    ret=$?
fi

# 随机地按某种方式启动，直到以某种方式正常启动以止；否则杀掉mysqld进程，随机休息一会儿，重试
while [ $ret -ne 0 ]; do
    kill_mysqld_process
    sleep_time=$(( $RANDOM % 10 ))
    sleep $sleep_time
    choice=$(( $RANDOM % 2 ))
    ret=-1
    if [ $choice -eq 0 ] ; then
        start_mariadb_special
        ret=$?
    else
        start_mariadb_normal
        ret=$?
    fi
done

# 使上述脚本开机自启动
chmod +x /usr/local/bin/mariadb_cluster_helper.sh
chmod +x /etc/rc.d/rc.local
echo '
/usr/local/bin/mariadb_cluster_helper.sh &> /var/log/mariadb_cluster_helper.log &' >> /etc/rc.d/rc.local

GRANT PROCESS ON *.* TO 'clustercheckuser'@'%' IDENTIFIED BY 'clustercheckpassword!'

wget -O /usr/bin/clustercheck https://raw.githubusercontent.com/olafz/percona-clustercheck/master/clustercheck
chmod +x /usr/bin/clustercheck

# /usr/bin/clustercheck > /dev/null
# echo $?
0 # synced
1 # un-synced

# 假设已经构建了一个两节点的galera集群，在第3个节点部署garbd服务
echo '
GALERA_NODES="10.211.55.6:4567 10.211.55.7:4567" # 这里是两节点的地址
GALERA_GROUP="galera_cluster"  # 这里的group名称保持与两节点的wsrep_cluster_name属性一致
LOG_FILE="/var/log/garb.log"
' > /etc/sysconfig/garb
systemctl start garb # 启动garbd服务

# 首先在第3个节点停止garb服务
systemctl stop garb

# 然后在第2个节点drop掉去住第1个节点和仲裁节点的数据包
iptables -A OUTPUT -d 10.211.55.6 -j DROP
iptables -A OUTPUT -d 10.211.55.9 -j DROP

# 这时检查前两个节点的同步状态，发生脑裂了，都不是同步状态了
mysql -e "show status like 'wsrep_local_state_comment'"
+---------------------------+-------------+
| Variable_name             | Value       |
+---------------------------+-------------+
| wsrep_local_state_comment | Initialized |
+---------------------------+-------------+

# 首先在第3个节点启动garb服务
systemctl start garb

# 在前两个节点查看集群节点数，发现是3个，说明包括了仲裁节点
mysql -e "show status like 'wsrep_cluster_size'"
+---------------------------+-------------+
| Variable_name             | Value       |
+---------------------------+-------------+
| wsrep_cluster_size        | 3           |
+---------------------------+-------------+

# 然后在第2个节点drop掉去住第1个节点和仲裁节点的数据包
iptables -A OUTPUT -d 10.211.55.6 -j DROP
iptables -A OUTPUT -d 10.211.55.9 -j DROP

# 这时检查第1个节点的同步状态，仍然是同步状态
mysql -e "show status like 'wsrep_local_state_comment'"
+---------------------------+-------------+
| Variable_name             | Value       |
+---------------------------+-------------+
| wsrep_local_state_comment | Synced      |
+---------------------------+-------------+

# 再在第1个节点查看集群节点数，发现是2个
mysql -e "show status like 'wsrep_cluster_size'"
+---------------------------+-------------+
| Variable_name             | Value       |
+---------------------------+-------------+
| wsrep_cluster_size        | 2           |
+---------------------------+-------------+

# 这时检查第2个节点的同步状态，发现是未同步的
mysql -e "show status like 'wsrep_local_state_comment'"
+---------------------------+-------------+
| Variable_name             | Value       |
+---------------------------+-------------+
| wsrep_local_state_comment | Initialized |
+---------------------------+-------------+