SUSE 11 SP3下搭建PXC(Percona XtraDB Cluster)
1. 下载源码
Percona-XtraDB-Cluster-5.7.19-29.22.tar.gz - https://www.percona.com/downloads/Percona-XtraDB-Cluster-LATEST/
openssl-1.2.j.tar.gz
2. 编译openssl
#CFLAGS=-fPIC ./config --prefix=/opt/openssl-1.0.2j/
#make && make install
3. 编译PXC
#mkdir output
#./build-ps/build-binary.sh output --with-ssl /opt/openssl-1.0.2j/
一切顺利的话，在output中可以找到最终编译结果的压缩包
4. 安装
#将编译出来的压缩包，解压到/opt/mysql-pxc
#cp support-files/mysql.server /etc/init.d/mysql-pxc
#useradd mysql ; groupadd mysql; usermod mysql -g mysql
#mkdir -p /data/mysql-pxc/{tmp,data,logs} ; chown -R mysql:mysql /data/mysql-pxc
#修改/etc/init.d/mysql-pxc文件
# 1) 注释掉. /etc/rc.d/init.d/functions
# 2) 修改basedir指向/opt/mysql-pxc
# 3) 修改datadir指向/data/mysql-pxc/data/
#修改/opt/mysql-pxc/bin/mysqld_safe中所有的硬编码路径，使指向/opt/mysql-pxc
5. 配置：3台服务器场景
服务器地址：157.0.0.51 , 157.0.0.52, 157.0.0.61
5.1 配置/etc/my.cnf
[mysqld]
server_id = 51# 必须保证每台服务器的编号不同
basedir = /opt/mysql-pxc
port = 3306
datadir = /data/mysql-pxc/data/
tmpdir = /data/mysql-pxc/tmp
socket = /data/mysql-pxc/tmp/mysql.sock
user = mysql
pid-file = /data/mysql-pxc/data/mysqld.pid
log-error = /data/mysql-pxc/logs/mysql.err
general-log-file = /data/mysql-pxc/logs/mysql.log
slow-query-log-file = /data/mysql-pxc/logs/mysql.log
default_storage_engine = InnoDB
innodb_locks_unsafe_for_binlog = 1
innodb_autoinc_lock_mode = 2
wsrep_cluster_name = "mycluster"
wsrep_cluster_address = gcomm://157.0.0.51,157.0.0.52,157.0.0.61
wsrep_node_address = 157.0.0.51 # 本机地址
wsrep_provider = /opt/mysql-pxc/lib/libgalera_smm.so
wsrep_sst_auth = "mytest:mytest"
wsrep_node_name = pxc51 # 本机名称
5.2 初始化 & 启动
1)初始化每台数据库
#/opt/mysql-pxc/bin/mysqld --initialize --basedir=/opt/mysql-pxc
2) 启动第一台
#/etc/init.d/mysql-pxc bootstrap-pxc --wsrep-cluster-address="gcomm://"
3)启动其它机器
#/etc/init.d/mysql-pxc start
4)重启第一台机器
#/etc/init.d/mysql-pxc restart
5)验证
#修改密码
#在mysql日志中查找初始密码，登录mysql: /opt/mysql-pxc/bin/mysql -uroot -p -h127.0.0.1
mysql>set password=password('root');
mysql>flush privileges;
# 查看集群节点个数
mysql> show global status like 'wsrep_cluster_size';
# 查看集群各节点地址
mysql> show global status like 'wsrep_incoming_addresses';
6. 补充
6.1 支持整个集群的binlog日志远程复制
选定一台PXC节点作为master，假设IP为157.0.0.51，假设远程slave机器地址为157.0.0.99，需要进行如下配置
1) 在每个PXC节点的my.cnf 配置文件[mysqld]节中增加如下几行配置
server_id = xx #每个节点不能相同
log-bin = mysql-bin
binlog-do-db = n1 # 需要复制的数据库名
replicate-do-db = n1 # 需要复制的数据库名
log_slave_updates
在PXC节点上增加一个用户例如：
mysql>grant replication slave on *.* to 'rep'@'157.0.0.99' identified by 'pass';
mysql>flush privileges;
2) 在远程的slave节点上配置
mysql>change master to master_host='157.0.0.51', master_user='rep', master_password='pass', master_log_file='mysql-bin.000001',master_log_pos=1;
mysql>start slave;
mysql>show slave status\G;
6.2 如何自举
重启mysql方式：
/etc/init.d/mysql-pxc bootstrap-pxc --wsrep-cluster-address="gcomm://"
不重启mysql方式：
mysql>SET GLOBAL wsrep_provider_options='pc.bootstrap=YES';
6.3 如何忽略脑裂
mysql>SET GLOBAL wsrep_provider_options="pc.ignore_sb=true";
6.4 设置节点权重，用于控制投票，可以解决脑裂问题，但同时会导致可用性降低
#默认都为1
mysql>SET GLOBAL wsrep_provider_options="pc.weight=3";
6.5 两节点避免脑裂的正确方法是使用PXC自带的仲裁者garbd