MHA构建MySQL高可用平台最佳实践

图片来自网络

文/Bruce.Liu1

文章大纲

  1. MHA简介
    1.1. mha组件介绍
    1.2. 背景和目标
  2. MHA原理
    2.1. MHA工作原理
    2.2. MHA工具介绍
    2.3. 当前高可用方案
    2.4. MHA的优势
  3. MHA最佳实践
    3.1. 背景介绍
    3.2. 安装MySQL实例
    3.3. 部署MySQL复制
    3.4. 部署MHA软件
    3.5. 故障自动切换与在线切换

1.MHA简介

MHA是什么?
MHA是由日本Mysql yoshinorim专家(原就职于DeNA现就职于FaceBook)用Perl写的一套Mysql故障切换方案,来保障数据库的高可用性,它的功能是能在0-30s之内实现主Mysql故障转移(failover),MHA故障转移可以很好的帮我们解决从库数据的一致性问题,同时最大化挽回故障发生后数据的一致性。
官方网站:https://code.google.com/p/mysql-master-ha/

MHA(Master High Availability)目前在MySQL高可用方面是一个相对成熟的解决方案,它由日本DeNA公司youshimaton(现就职于Facebook公司)开发,是一套优秀的作为MySQL高可用性环境下故障切换和主从提升的高可用软件。在MySQL故障切换过程中,MHA能做到在0~30秒之内自动完成数据库的故障切换操作,并且在进行故障切换的过程中,MHA能在较大程度上保证数据的一致性,以达到真正意义上的高可用。

该软件由两部分组成:MHA Manager(管理节点)和MHA Node(数据节点)。MHA Manager可以单独部署在一台独立的机器上管理多个master-slave集群,也可以部署在一台slave节点上。MHA Node运行在每台MySQL服务器上,MHA Manager会定时探测集群中的master节点,当master出现故障时,它可以自动将数据的slave提升为新的master,然后将所有其他的slave重新指向新的master。整个故障转移过程对应用程序完全透明。

在MHA自动故障切换过程中,MHA试图从宕机的主服务器上保存二进制日志,较大程度的保证数据的不丢失,但这并不总是可行的。例如,如果主服务器硬件故障或无法通过ssh访问,MHA没法保存二进制日志,只进行故障转移而丢失了的数据。使用MySQL 5.5的半同步复制,可以大大降低数据丢失的风险。MHA可以与半同步复制结合起来。如果只有一个slave已经收到了的二进制日志,MHA可以将的二进制日志应用于其他所有的slave服务器上,因此可以保证所有节点的数据一致性。

1.1.mha组件介绍

  • MHA Manager
    运行一些工具,比如masterha_manager工具实现自动监控MySQL Master和实现master故障切换,其它工具实现手动实现master故障切换、在线mater转移、连接检查等等。一个Manager可以管理多 个master-slave集群

  • MHA Node
    部署在所有运行MySQL的服务器上,无论是master还是slave。主要作用有三个。
    1.保存二进制日志
    如果能够访问故障master,会拷贝master的二进制日志
    2.应用差异中继日志
    从拥有最新数据的slave上生成差异中继日志,然后应用差异日志。
    3.清除中继日志
    在不停止SQL线程的情况下删除中继日志

1.2.背景和目标

在早期的MySQL架构中最主流就就是MySQL复制的主从结构,但伴随时间的推移以及数据的膨胀会出现一下几类问题。

  • 以前几十台DB服务器,人工登陆服务器就能维护好,也没有高可用,当master挂了,通知业务将IP切换到slave然后重启也能基本满足业务要求,但是业务迅速发展,实例数不断增加,复制集不断增加,数据库架构多样化,而这种人工维护方式显然大大增加了DBA工作量,而且效率低下、容易出错。

  • DB规模的增大,机器故障、SQL故障、实例故障出现的概率也增加、还有来自业务方的DB变更,比如大表增加字段、增加索引、批量删除数据等异常维护操作,当然这些在一定条件下可用采用在线变更,比如采用pt-online-schema-change工具,但是当不满足在线变更条件、或者在线变更复杂的情况下,就需要采用滚动变更的方式,先在各个slave上变更、在线切换后再在master上变更,然后再进行一次切换还原,而这些切换操作如果全部手工敲命令来进行显然是不可取的。

  • 随着用户数的不断增加,业务方对DB这种基础服务的可用性也就越来越高,在魅族业务对DB的可用性要求是每个月需要达到四个9,也就意味着每个月的故障时间只有不到5分钟,以前那种通知业务更改IP重启的方式显然是达不到这个要求的。

    在这些背景和要求下,我们需要摆脱手工操作,需要一套有效的MySQL高可用方案和一个高效的高可用平台来支撑DB的快速增长。MySQL高可用平台需要达到的目标有以下几点:

    1.数据一致性保证这个是最基本的同时也是前提,如果主备的数据的不一致,那么切换就无法进行,当然这里的一致性也是一个相对的,但是要做到最终一致性。
    2.故障快速切换,当master故障时这里可以是机器故障或者是实例故障,要确保业务能在最短时间切换到备用节点,使得业务受影响时间最短。这里也可以指业务例行维护操作,比如前面提到的无法使用在线进行DDL的DDL操作,很多分表批量的DDL操作,这些操作通过在线切换方式来滚动完成。
    3.简化日常维护,通过高可用平台来自动完成高可用的部署、维护、监控等任务,能够最大程度的解放DBA手动操作,提高日常运维效率。
    4.统一管理,当复制集很多的情况下,能够统一管理高可用实例信息、实例信息、监控信息、切换信息等。
    高可用的部署要对现有的数据库架构无影响,如果因为部署高可用,需要更改或者调整数据库架构则会导致成本增加。

2.MHA原理

2.1.MHA工作原理

image.png

当master出现故障时,通过对比slave之间I/O线程读取masterbinlog的位置,选取最接近的slave做为latestslave。 其它slave通过与latest slave对比生成差异中继日志。在latest slave上应用从master保存的binlog,同时将latest slave提升为master。最后在其它slave上应用相应的差异中继日志并开始从新的master开始复制。

在MHA实现Master故障切换过程中,MHA Node会试图访问故障的master(通过SSH),如果可以访问(不是硬件故障,比如InnoDB数据文件损坏等),会保存二进制文件,以最大程度保 证数据不丢失。MHA和半同步复制一起使用会大大降低数据丢失的危险。流程如下:

  • 从宕机崩溃的master保存二进制日志事件(binlog events)。
  • 识别含有最新更新的slave。
  • 应用差异的中继日志(relay log)到其它slave。
  • 应用从master保存的二进制日志事件(binlog events)。
  • 提升一个slave为新master并记录binlog file和position。
  • 使其它的slave连接新的master进行复制。
  • 完成切换manager主进程OFFLINE

2.2.MHA工具介绍

1.Manager工具:

  • masterha_check_ssh : 检查MHA的SSH配置。
  • masterha_check_repl : 检查MySQL复制。
  • masterha_manager : 启动MHA。
  • masterha_check_status : 检测当前MHA运行状态。
  • masterha_master_monitor : 监测master是否宕机。
  • masterha_master_switch : 控制故障转移(自动或手动)。
  • masterha_conf_host : 添加或删除配置的server信息。

2. Node工具

  • save_binary_logs : 保存和复制master的二进制日志。
  • apply_diff_relay_logs : 识别差异的中继日志事件并应用于其它slave。
  • filter_mysqlbinlog : 去除不必要的ROLLBACK事件(MHA已不再使用这个工具)。
  • purge_relay_logs : 清除中继日志(不会阻塞SQL线程)。
    注意:Node这些工具通常由MHA Manager的脚本触发,无需人手操作。

2.3.当前高可用方案

  • keepalived+mysql复制
    该结构与MHA类似,但keepalived的优势在于无状态组件的故障切换,常用于web前端的故障转移,应用于数据库场景中,最致命的问题就是脑裂以后数据乱写的风险,为企业带来巨大困扰。

  • MySQL Cluster
    MySQL Cluster真正实现了高可用,但是使用的是NDB存储引擎,并且SQL节点有单点故障问题。

  • 半同步复制(5.5+)
    半同步复制大大减少了“binlog events只存在故障master上”的问题。在提交时,保证至少一个slave(并不是所有的)接收到binlog,因此一些slave可能没有接收到binlog。

  • PXC
    PXC实现了服务高可用,数据同步时是并发复制。但是仅支持InnoDB引擎,所有的表都要有主键。锁冲突、死锁问题相对较多等等问题。

2.4.MHA的优势

  • 障切换快
    在 主从复制集群中,只要从库在复制上没有延迟,MHA通常可以在数秒内实现故障切换。9-10秒内检查到master故障,可以选择在7-10秒关闭 master以避免出现裂脑,几秒钟内,将差异中继日志(relay log)应用到新的master上,因此总的宕机时间通常为10-30秒。恢复新的master后,MHA并行的恢复其余的slave。即使在有数万台 slave,也不会影响master的恢复时间。
    DeNA在超过150个MySQL(主要5.0/5.1版本)主从环境下使用了MHA。当mater故障后,MHA在4秒内就完成了故障切换。在传统的主动/被动集群解决方案中,4秒内完成故障切换是不可能的。

  • master故障不会导致数据不一致
    当 目前的master出现故障是,MHA自动识别slave之间中继日志(relay log)的不同,并应用到所有的slave中。这样所有的salve能够保持同步,只要所有的slave处于存活状态。和Semi- Synchronous Replication一起使用,(几乎)可以保证没有数据丢失。

  • 需修改当前的MySQL设置
    MHA的设计的重要原则之一就是尽可能地简单易用。MHA工作在传统的MySQL版本5.0和之后版本的主从复制环境中。和其它高可用解决方法比,MHA并不需要改变MySQL的部署环境。MHA适用于异步和半同步的主从复制。
    启动/停止/升级/降级/安装/卸载MHA不需要改变(包扩启动/停止)MySQL复制。当需要升级MHA到新的版本,不需要停止MySQL,仅仅替换到新版本的MHA,然后重启MHA Manager就好了。
    MHA运行在MySQL 5.0开始的原生版本上。一些其它的MySQL高可用解决方案需要特定的版本(比如MySQL集群、带全局事务ID的MySQL等等),但并不仅仅为了 master的高可用才迁移应用的。在大多数情况下,已经部署了比较旧MySQL应用,并且不想仅仅为了实现Master的高可用,花太多的时间迁移到不 同的存储引擎或更新的前沿发行版。MHA工作的包括5.0/5.1/5.5的原生版本的MySQL上,所以并不需要迁移。

  • 无需增加大量的服务器
    MHA由MHA Manager和MHA Node组成。MHA Node运行在需要故障切换/恢复的MySQL服务器上,因此并不需要额外增加服务器。MHA Manager运行在特定的服务器上,因此需要增加一台(实现高可用需要2台),但是MHA Manager可以监控大量(甚至上百台)单独的master,因此,并不需要增加大量的服务器。即使在一台slave上运行MHA Manager也是可以的。综上,实现MHA并没用额外增加大量的服务。

  • 无性能下降
    MHA适用与异步或半同步的MySQL复制。监控master时,MHA仅仅是每隔几秒(默认是3秒)发送一个ping包,并不发送重查询。可以得到像原生MySQL复制一样快的性能。

  • 适用于任何存储引擎
    MHA可以运行在只要MySQL复制运行的存储引擎上,并不仅限制于InnoDB,即使在不易迁移的传统的MyISAM引擎环境,一样可以使用MHA。

3.MHA最佳实践

图片来自网络

3.1.背景介绍

3.1.1.软件参考文档

参考文档:
MHA原理:https://code.google.com/p/mysql-master-ha/wiki/HowMHAWorks
MHA原理PPT:http://www.slideshare.net/matsunobu/automated-master-failover
Linux配置代理方法:http://blog.csdn.net/bojie5744/article/details/42148719

软件下载:
Centos Base Yum Repository: http://mirrors.163.com/.help/CentOS6-Base-163.repo
epel(RHEL 6)Yum Repository:http://dl.fedoraproject.org/pub/epel/6/x86_64/epel-release-6-8.noarch.rpm
epel(RHEL 7)Yum Repository:
https://mirrors.tuna.tsinghua.edu.cn/epel//7/x86_64/Packages/e/epel-release-7-11.noarch.rpm
MySQL5.7 Yum Repository:https://dev.mysql.com/get/mysql57-community-release-el6-11.noarch.rpm
mysql-master-ha(mgr):https://github.com/linyue515/mysql-master-ha/raw/master/mha4mysql-manager-0.57-0.el7.noarch.rpm
mysql-master-ha(node):https://github.com/linyue515/mysql-master-ha/raw/master/mha4mysql-node-0.57-0.el7.noarch.rpm

3.1.2.系统环境介绍
图片来自原创
  • 系统版本
    CentOS release 6.7 (Final) x86_64

  • MySQL版本
    mysql-5.7.20.-x86_64(RPM)

  • MHA版本
    mha4mysql-manager-0.57
    mha4mysql-node-0.57

3.1.3.安装系统要求
  • 涉及所有服务器关闭iptables、NetworkManager服务、selinux安全配置
# /etc/init.d/NetworkManager stop
# chkconfig NetworkManager off
# /etc/init.d/iptables stop
# chkconfig iptables off

/etc/selinux/config 改成disable

3.2.安装MySQL实例

3.2.1.安装mysql数据库
  • 创建mysql用户
# useradd mysql
# passwd mysql
  • 安装软件
# yum -y install mysql-community-server.x86_64
3.2.2.创建配置文件目录
# mkdir /etc/mysql
3.2.3.创建配置文件
[mysqld]
# GENERAL #
user                           = mysql
port                           = 3389
default_storage_engine         = InnoDB
socket                         = /data1/db3389/my3389.sock
pid_file                       = /data1/db3389/mysql.pid
#read-only =0
tmpdir                  = /data1/tmp
#key_buffer_size                = 128M
max_allowed_packet             = 32M
max_connect_errors             = 1000000
datadir          = /data1/db3389/
log_bin = 2171303389-bin
relay-log=  2171303389-relay-bin
expire_logs_days               = 7
#sync_binlog                    = 0
tmp_table_size                 = 32M
max_heap_table_size            = 32M
max_connections                = 5000
thread_cache_size              = 512
table_definition_cache         = 4096
table_open_cache               = 4096
wait_timeout            = 28800
interactive_timeout     = 28800
transaction-isolation = READ-COMMITTED
binlog-format=row
character-set-server=utf8
skip-name-resolve
back_log=1024
explicit_defaults_for_timestamp=true
server_id=2171303389

# INNODB #
innodb_flush_method            = O_DIRECT
#innodb_data_home_dir = /data1/db3389
innodb_data_file_path = ibdata1:100M:autoextend
#redo log
#innodb_log_group_home_dir=./
innodb_log_files_in_group      = 3
innodb_log_file_size           = 128M
#innodb performance
innodb_flush_log_at_trx_commit = 0
innodb_file_per_table          = 1
innodb_buffer_pool_instances   = 8
innodb_io_capacity             = 2000
innodb_lock_wait_timeout       = 30
binlog_error_action = ABORT_SERVER
innodb_buffer_pool_size        = 128M
innodb_max_dirty_pages_pct=90
innodb_file_format=Barracuda
innodb_support_xa=0
innodb_buffer_pool_dump_at_shutdown = 1
innodb_buffer_pool_load_at_startup = 1
#innodb undo log
innodb_undo_tablespaces=4
innodb_undo_logs=2048
innodb_purge_rseg_truncate_frequency=512
innodb_max_undo_log_size=2G
innodb_undo_log_truncate=1

log_error                      = error.log
#log_queries_not_using_indexes = 1
slow_query_log                 = 1
slow_query_log_file            = slow-queries.log
long_query_time=2
gtid_mode=ON
enforce-gtid-consistency
log-slave-updates
master-info-repository=TABLE
relay-log-info-repository=TABLE
sync_master_info = 10000
slave_sql_verify_checksum=1
skip-slave-start
init-connect='SET NAMES utf8'
character-set-server=utf8
skip-character-set-client-handshake
bind-address=0.0.0.0
skip-external-locking
slave-parallel-workers=6

[mysql5.6]
myisam_recover                 = FORCE,BACKUP
3.2.4.创建授权目录
# mkdir -p /data1/db3389
# mkdir -p /data1/tmp
# chown -R mysql:mysql /data1/db3389
# chown -R mysql:mysql /data1/tmp
3.2.5.初始化MySQL实例
# mysqld --defaults-file=/etc/mysql/my3389.cnf --initialize --user=mysql
# mysql_ssl_rsa_setup 
3.2.6.启动mysql实例
# mysqld_safe --defaults-file=/etc/mysql/my3389.cnf &
# cat /data1/db3389/error.log | grep temp
# mysql -S /data1/db3389/my3389.sock -p'z&Di4b_oSM*-'
mysql> set password=''; #重置密码为空

3.3.部署MySQL复制

3.3.1.主库创建复制用户
mysql> grant replication slave, replication client on *.* to replica@'192.168.217.%' identified by 'mycatDBA';
3.3.2.主库创建mha用户
mysql> grant all privileges on *.* to mha@'192.168.217.132' identified by 'mysqlDBA';
3.3.3.主库备份数据库
# mysqldump -S /data1/db3389/my3389.sock --single-transaction --master-data=2 --opt -A | gzip >  /data1/tmp/full_3389.tar.gz
3.3.4.主库传输至从库
# scp /data1/tmp/full_3389.tar.gz 192.168.217.131:/data1/tmp
3.3.5.从库恢复数据库
# gunzip < /data1/tmp/full_3389.tar.gz | mysql -S /data1/db3389/my3389.sock

注意:恢复数据库前,从库最好reset master;,否则将出现一下错误:
ERROR 1840 (HY000) at line 24: @@GLOBAL.GTID_PURGED can only be set when @@GLOBAL.GTID_EXECUTED is empty.

3.3.6.从库初始化同步数据
mysql> change master to master_host='192.168.217.130',master_port=3389,master_user='replica',master_password='mycatDBA',master_auto_position=1;
Query OK, 0 rows affected, 2 warnings (0.02 sec)

mysql> start slave;
Query OK, 0 rows affected (0.03 sec)


mysql> show slave status \G
*************************** 1. row ***************************
...... 省略 ......
             Slave_IO_Running: Yes
            Slave_SQL_Running: Yes
...... 省略 ......

3.4.部署MHA软件

3.4.1.安装软件
  • epel yum源安装方式
# yum -y install perl-DBD-MySQL perl-Config-Tiny perl-Log-Dispatch perl-Parallel-ForkManager perl-Time-HiRes
# #根据MHA角色安装对应的软件包即可
# yum -y --nogpgcheck install mha4mysql-node-0.57-0.el7.noarch.rpm
# yum -y install --nogpgcheck mha4mysql-manager-0.57-0.el7.noarch.rpm
  • 本地安装方式
# yum -y --nogpgcheck install perl-DBD-MySQL*
# yum -y --nogpgcheck install perl-Config-Tiny*
# yum -y --nogpgcheck install perl-Parallel-ForkManager*
# yum -y --nogpgcheck install  perl-MailTools*
# yum -y --nogpgcheck install perl-Email-Date-Format*
# yum -y --nogpgcheck install perl-Mail-Sender*
# yum -y --nogpgcheck install perl-MIME-Types*
# yum -y --nogpgcheck install perl-MIME-Lite*
# yum -y --nogpgcheck install perl-Mail-Sendmail*
# yum -y --nogpgcheck install perl-Log-Dispatch*
# #根据MHA角色安装对应的软件包即可 
# yum -y --nogpgcheck install mha4mysql-node-0.57-0.el7.noarch.rpm
# yum -y install --nogpgcheck mha4mysql-manager-0.57-0.el7.noarch.rpm
3.4.2.挂在VIP
  • master
# /sbin/ifconfig eth0:1 192.168.217.201 broadcast 192.168.217.255 netmask 255.255.255.0
# /sbin/arping -f -q -c 5 -w 5 -I eth0 -s 192.168.217.201 -U 192.168.217.2
3.4.3.配置SSH互信

在现网环境中几乎都是禁止root远程登陆服务器得,所以ssh免密码登陆要在mysql用户下进行配置,这是处于安全角度考虑出发。

  • master:
# su - mysql
$ ssh-keygen -t rsa
$ rm -rf ~/.ssh/*
  • slave:
# su - mysql
$ ssh-keygen -t rsa
$ rm -rf ~/.ssh/*
  • manager:
# su - mysql
$ ssh-keygen -t rsa
$ cd ~/.ssh
$ mv id_rsa.pub authorized_keys
$ scp * 192.168.217.130:~/.ssh/
$ scp * 192.168.217.131:~/.ssh/
$ #测试ssh
$ ssh 192.168.217.130 date 
Wed Nov 22 05:48:54 PST 2017
$ ssh 192.168.217.131 date 
Wed Nov 22 05:47:58 PST 2017
3.4.4.配置mysql用户sudo权限
  • 添加普通用户登陆tty终端权限
# vim /etc/sudoers

#将以下的参数注释,意思就是sudo默认需要tty终端。注释掉就可以在后台执行了。
#Defaults    requiretty
  • 开放普通用户执行sudo命令权限
# cd /etc/sudoers.d/
# vim mysql

User_Alias  MYSQL_USERS = ALL
Runas_Alias MYSQL_RUNAS = root
Cmnd_Alias  MYSQL_CMNDS = ALL
MYSQL_USERS ALL = (MYSQL_RUNAS) NOPASSWD: MYSQL_CMNDS
3.4.5.创建MHA配置文件
  • 创建配置文件目录
# mkdir /etc/mha
  • 创建MHA配置文件
# cat app3389.cnf 
[server default]
user=mha
password=mysqlDBA
manager_workdir=/data1/mha/masterha/app3389
manager_log=/data1/mha/masterha/app3389/app3389.log
remote_workdir=/data1/mha/masterha/app3389
ssh_user=mysql
repl_user=replica    
repl_password=mycatDBA
ping_interval=3         

secondary_check_script="masterha_secondary_check -s 192.168.1.122 -s 192.168.1.122"
master_ip_failover_script="/etc/mha/master_ip_failover.sh 192.168.1.201 1"
master_ip_online_change_script="/etc/mha/master_ip_online_change.sh 192.168.1.201 1"
shutdown_script="/etc/mha/power_manager"
#report_script="/etc/mha/end_report"

[server1]
hostname=192.168.1.120
port=3389
master_binlog_dir=/data1/db3389
candidate_master=1   
master_pid_file=/data1/db3389/mysql.pid               

[server2]
hostname=192.168.1.121
port=3389
master_binlog_dir=/data1/db3389
candidate_master=1
master_pid_file=/data1/db3389/mysql.pid    

[binlog1]
hostname=192.168.1.122
master_binlog_dir=/data1/mha/binlog/3389
no_master=1
ignore_fail=1
3.4.6.上传MHA切换脚本

master_ip_failover.sh
master_ip_online_change.sh
power_manager

注意:脚本内容中要修改网卡名字

my $vip  = shift;
my $interface = 'eth1';
my $key = shift;
  • 上传故障切换脚本并授权
# chmod 755 master_ip_*
# chmod 755 power_manager
3.4.7.创建MHA、BINLOG工作目录
# mkdir -p /data1/mha/masterha/app3389
# mkdir -p /data1/mha/binlog/3389
# chown -R mysql:mysql /data1/mha/binlog/3389
3.4.8.启动BINLOG SERVER
# su - mysql
$ cd /data1/mha/binlog/3389;
$ mysqlbinlog -R --host=192.168.217.130 -P3389 --user=mha --password=mysqlDBA  --raw --stop-never 2171303389-bin.000003 &
$ ps -ef | grep mysqlbinlog | grep -v grep  # 验证binlog server进程是否存在
root       7008   6233  0 07:00 pts/0    00:00:00 mysqlbinlog -R --host=192.168.217.130 -P3389 --user=mha --password=x xxxxxx --raw --stop-never 2171303389-bin.000003
3.4.9.验证并启动manager进程
$ masterha_check_ssh --conf=/etc/mha/app3389.cnf 
Wed Nov 22 07:35:07 2017 - [warning] Global configuration file /etc/masterha_default.cnf not found. Skipping.
Wed Nov 22 07:35:07 2017 - [info] Reading application default configuration from /etc/mha/app3389.cnf..
Wed Nov 22 07:35:07 2017 - [info] Reading server configuration from /etc/mha/app3389.cnf..
Wed Nov 22 07:35:07 2017 - [info] Starting SSH connection tests..
Wed Nov 22 07:35:08 2017 - [debug] 
Wed Nov 22 07:35:07 2017 - [debug]  Connecting via SSH from root@192.168.217.130(192.168.217.130:22) to root@192.168.217.131(192.168.217.131:22)..
Wed Nov 22 07:35:08 2017 - [debug]   ok.
Wed Nov 22 07:35:08 2017 - [debug] 
Wed Nov 22 07:35:07 2017 - [debug]  Connecting via SSH from root@192.168.217.131(192.168.217.131:22) to root@192.168.217.130(192.168.217.130:22)..
Wed Nov 22 07:35:08 2017 - [debug]   ok.
Wed Nov 22 07:35:08 2017 - [info] All SSH connection tests passed successfully.
$ masterha_check_repl --conf=/etc/mha/app3389.cnf 
Wed Nov 22 07:47:07 2017 - [warning] Global configuration file /etc/masterha_default.cnf not found. Skipping.
Wed Nov 22 07:47:07 2017 - [info] Reading application default configuration from /etc/mha/app3389.cnf..
Wed Nov 22 07:47:07 2017 - [info] Reading server configuration from /etc/mha/app3389.cnf..
Wed Nov 22 07:47:07 2017 - [info] MHA::MasterMonitor version 0.57.
Wed Nov 22 07:47:08 2017 - [info] GTID failover mode = 1
Wed Nov 22 07:47:08 2017 - [info] Dead Servers:
Wed Nov 22 07:47:08 2017 - [info] Alive Servers:
Wed Nov 22 07:47:08 2017 - [info]   192.168.217.130(192.168.217.130:3389)
Wed Nov 22 07:47:08 2017 - [info]   192.168.217.131(192.168.217.131:3389)
Wed Nov 22 07:47:08 2017 - [info] Alive Slaves:
Wed Nov 22 07:47:08 2017 - [info]   192.168.217.131(192.168.217.131:3389)  Version=5.7.20-log (oldest major version between slaves) log-bin:enabled
Wed Nov 22 07:47:08 2017 - [info]     GTID ON
Wed Nov 22 07:47:08 2017 - [info]     Replicating from 192.168.217.130(192.168.217.130:3389)
Wed Nov 22 07:47:08 2017 - [info]     Primary candidate for the new Master (candidate_master is set)
Wed Nov 22 07:47:08 2017 - [info] Current Alive Master: 192.168.217.130(192.168.217.130:3389)
Wed Nov 22 07:47:08 2017 - [info] Checking slave configurations..
Wed Nov 22 07:47:08 2017 - [info]  read_only=1 is not set on slave 192.168.217.131(192.168.217.131:3389).
Wed Nov 22 07:47:08 2017 - [info] Checking replication filtering settings..
Wed Nov 22 07:47:08 2017 - [info]  binlog_do_db= , binlog_ignore_db= 
Wed Nov 22 07:47:08 2017 - [info]  Replication filtering check ok.
Wed Nov 22 07:47:08 2017 - [info] GTID (with auto-pos) is supported. Skipping all SSH and Node package checking.
Warning: Permanently added '192.168.217.132' (RSA) to the list of known hosts.
Wed Nov 22 07:47:08 2017 - [info] HealthCheck: SSH to 192.168.217.132 is reachable.
Wed Nov 22 07:47:14 2017 - [info] Binlog server 192.168.217.132 is reachable.
Wed Nov 22 07:47:14 2017 - [info] Checking recovery script configurations on 192.168.217.132(192.168.217.132:3306)..
Wed Nov 22 07:47:14 2017 - [info]   Executing command: save_binary_logs --command=test --start_pos=4 --binlog_dir=/data1/mha/binlog/3389 --output_file=/data1/mha/masterha/app3389/save_binary_logs_test --manager_version=0.57 --start_file=2171303389-bin.000003 
Wed Nov 22 07:47:14 2017 - [info]   Connecting to root@192.168.217.132(192.168.217.132:22).. 
  Creating /data1/mha/masterha/app3389 if not exists..    ok.
  Checking output directory is accessible or not..
   ok.
  Binlog found at /data1/mha/binlog/3389, up to 2171303389-bin.000003
Wed Nov 22 07:47:14 2017 - [info] Binlog setting check done.
Wed Nov 22 07:47:14 2017 - [info] Checking SSH publickey authentication settings on the current master..
Wed Nov 22 07:47:15 2017 - [info] HealthCheck: SSH to 192.168.217.130 is reachable.
Wed Nov 22 07:47:15 2017 - [info] 
192.168.217.130(192.168.217.130:3389) (current master)
 +--192.168.217.131(192.168.217.131:3389)

Wed Nov 22 07:47:15 2017 - [info] Checking replication health on 192.168.217.131..
Wed Nov 22 07:47:15 2017 - [info]  ok.
Wed Nov 22 07:47:15 2017 - [info] Checking master_ip_failover_script status:
Wed Nov 22 07:47:15 2017 - [info]   /etc/mha/master_ip_failover.sh 192.168.217.201  1 --command=status --ssh_user=root --orig_master_host=192.168.217.130 --orig_master_ip=192.168.217.130 --orig_master_port=3389 
Checking the Status of the script.. OK 
Wed Nov 22 07:47:15 2017 - [info]  OK.
Wed Nov 22 07:47:15 2017 - [info] Checking shutdown script status:
Wed Nov 22 07:47:15 2017 - [info]   /etc/mha/power_manager --command=status --ssh_user=root --host=192.168.217.130 --ip=192.168.217.130 
Wed Nov 22 07:47:15 2017 - [info]  OK.
Wed Nov 22 07:47:15 2017 - [info] Got exit code 0 (Not master dead).

MySQL Replication Health is OK.
$ nohup masterha_manager --conf=/etc/mha/app3389.cnf --ignore_last_failover &
[2] 7307
$ nohup: ignoring input and appending output to `nohup.out'

$ masterha_check_status --conf=/etc/mha/app3389.cnf 
app3389 (pid:7307) is running(0:PING_OK), master:192.168.217.130

3.5.故障自动切换与在线切换

3.5.1.故障切换
  • 主库down或者主机down,然后测试切换是否成功。
3.5.2.在线切换

在线切换(Mha manager进程(binlog server进程可选)是关闭的,Mha结构是正常的环境,适用于生产系统硬件、软件升级维护等场景)

  • --orig_master_is_new_slave
    切换时加上此参数是讲原master变成slave节点,不加该参数,原master将不启动
  • --running_updates_limit=10000
    切换时选master 如果有延迟的话,mha切换不会成功,加上此参数表示切换在此时间范围内都可以切换(单位为 s),但是切换的时间长短是由recover时relay日志大小决定

注意:在备库先执行DDL,一般先stop slave,一般不记录mysql日志,可以通过set session sql_log_bin=0实现,然后进行一次主备切换操作,再在原来的主库上执行DDL.这种方法适用于增减索引.

$ masterha_master_switch --master_state=alive --conf=/etc/mha/app3389.conf --orig_master_is_new_slave
Sat Nov 25 11:06:04 2017 - [info] MHA::MasterRotate version 0.57.
Sat Nov 25 11:06:04 2017 - [info] Starting online master switch..
Sat Nov 25 11:06:04 2017 - [info] 
Sat Nov 25 11:06:04 2017 - [info] * Phase 1: Configuration Check Phase..
Sat Nov 25 11:06:04 2017 - [info] 
Sat Nov 25 11:06:04 2017 - [warning] Global configuration file /etc/masterha_default.cnf not found. Skipping.
Sat Nov 25 11:06:04 2017 - [info] Reading application default configuration from /etc/mha/app3389.conf..
Sat Nov 25 11:06:04 2017 - [info] Reading server configuration from /etc/mha/app3389.conf..
Sat Nov 25 11:06:04 2017 - [info] GTID failover mode = 1
Sat Nov 25 11:06:04 2017 - [info] Current Alive Master: 192.168.1.121(192.168.1.121:3389)
Sat Nov 25 11:06:04 2017 - [info] Alive Slaves:
Sat Nov 25 11:06:04 2017 - [info]   192.168.1.120(192.168.1.120:3389)  Version=5.7.20-log (oldest major version between slaves) log-bin:enabled
Sat Nov 25 11:06:04 2017 - [info]     GTID ON
Sat Nov 25 11:06:04 2017 - [info]     Replicating from 192.168.1.121(192.168.1.121:3389)
Sat Nov 25 11:06:04 2017 - [info]     Primary candidate for the new Master (candidate_master is set)

It is better to execute FLUSH NO_WRITE_TO_BINLOG TABLES on the master before switching. Is it ok to execute on 192.168.1.121(192.168.1.121:3389)? (YES/no): YES
Sat Nov 25 11:06:07 2017 - [info] Executing FLUSH NO_WRITE_TO_BINLOG TABLES. This may take long time..
Sat Nov 25 11:06:07 2017 - [info]  ok.
Sat Nov 25 11:06:07 2017 - [info] Checking MHA is not monitoring or doing failover..
Sat Nov 25 11:06:07 2017 - [info] Checking replication health on 192.168.1.120..
Sat Nov 25 11:06:07 2017 - [info]  ok.
Sat Nov 25 11:06:07 2017 - [info] Searching new master from slaves..
Sat Nov 25 11:06:07 2017 - [info]  Candidate masters from the configuration file:
Sat Nov 25 11:06:07 2017 - [info]   192.168.1.120(192.168.1.120:3389)  Version=5.7.20-log (oldest major version between slaves) log-bin:enabled
Sat Nov 25 11:06:07 2017 - [info]     GTID ON
Sat Nov 25 11:06:07 2017 - [info]     Replicating from 192.168.1.121(192.168.1.121:3389)
Sat Nov 25 11:06:07 2017 - [info]     Primary candidate for the new Master (candidate_master is set)
Sat Nov 25 11:06:07 2017 - [info]   192.168.1.121(192.168.1.121:3389)  Version=5.7.20-log log-bin:enabled
Sat Nov 25 11:06:07 2017 - [info]     GTID ON
Sat Nov 25 11:06:07 2017 - [info]  Non-candidate masters:
Sat Nov 25 11:06:07 2017 - [info]  Searching from candidate_master slaves which have received the latest relay log events..
Sat Nov 25 11:06:07 2017 - [info] 
From:
192.168.1.121(192.168.1.121:3389) (current master)
 +--192.168.1.120(192.168.1.120:3389)

To:
192.168.1.120(192.168.1.120:3389) (new master)
 +--192.168.1.121(192.168.1.121:3389)

Starting master switch from 192.168.1.121(192.168.1.121:3389) to 192.168.1.120(192.168.1.120:3389)? (yes/NO): YES
Sat Nov 25 11:06:11 2017 - [info] Checking whether 192.168.1.120(192.168.1.120:3389) is ok for the new master..
Sat Nov 25 11:06:11 2017 - [info]  ok.
Sat Nov 25 11:06:11 2017 - [info] 192.168.1.121(192.168.1.121:3389): SHOW SLAVE STATUS returned empty result. To check replication filtering rules, temporarily executing CHANGE MASTER to a dummy host.
Sat Nov 25 11:06:11 2017 - [info] 192.168.1.121(192.168.1.121:3389): Resetting slave pointing to the dummy host.
Sat Nov 25 11:06:11 2017 - [info] ** Phase 1: Configuration Check Phase completed.
Sat Nov 25 11:06:11 2017 - [info] 
Sat Nov 25 11:06:11 2017 - [info] * Phase 2: Rejecting updates Phase..
Sat Nov 25 11:06:11 2017 - [info] 
Sat Nov 25 11:06:11 2017 - [info] Executing master ip online change script to disable write on the current master:
Sat Nov 25 11:06:11 2017 - [info]   /etc/mha/master_ip_online_change.sh 192.168.1.200 1 --command=stop --orig_master_host=192.168.1.121 --orig_master_ip=192.168.1.121 --orig_master_port=3389 --orig_master_user='mha' --new_master_host=192.168.1.120 --new_master_ip=192.168.1.120 --new_master_port=3389 --new_master_user='mha' --orig_master_ssh_user=mysql --new_master_ssh_user=mysql   --orig_master_is_new_slave --orig_master_password=xxx --new_master_password=xxx
Unknown option: orig_master_ssh_user
Unknown option: new_master_ssh_user
Unknown option: orig_master_is_new_slave
Sat Nov 25 11:06:11 2017 918769 Set read_only on the new master.. ok.
Sat Nov 25 11:06:11 2017 923401 Waiting all running 1 threads are disconnected.. (max 1500 milliseconds)
{'Time' => '78','Command' => 'Binlog Dump GTID','db' => undef,'Id' => '46','Info' => undef,'User' => 'replica','State' => 'Master has sent all binlog to slave; waiting for more updates','Host' => '192.168.1.120:39100'}
Sat Nov 25 11:06:12 2017 426422 Waiting all running 1 threads are disconnected.. (max 1000 milliseconds)
{'Time' => '79','Command' => 'Binlog Dump GTID','db' => undef,'Id' => '46','Info' => undef,'User' => 'replica','State' => 'Master has sent all binlog to slave; waiting for more updates','Host' => '192.168.1.120:39100'}
Sat Nov 25 11:06:12 2017 929834 Waiting all running 1 threads are disconnected.. (max 500 milliseconds)
{'Time' => '79','Command' => 'Binlog Dump GTID','db' => undef,'Id' => '46','Info' => undef,'User' => 'replica','State' => 'Master has sent all binlog to slave; waiting for more updates','Host' => '192.168.1.120:39100'}
Sat Nov 25 11:06:13 2017 433392 Set read_only=1 on the orig master.. ok.
Sat Nov 25 11:06:13 2017 436292 Waiting all running 1 queries are disconnected.. (max 500 milliseconds)
{'Time' => '80','Command' => 'Binlog Dump GTID','db' => undef,'Id' => '46','Info' => undef,'User' => 'replica','State' => 'Master has sent all binlog to slave; waiting for more updates','Host' => '192.168.1.120:39100'}
Disabling the VIP on old master: 192.168.1.121 
===========sudo /sbin/ifconfig eth1:1 down===========================
Sat Nov 25 11:06:14 2017 071486 Killing all application threads..
Sat Nov 25 11:06:14 2017 072793 done.
Sat Nov 25 11:06:14 2017 - [info]  ok.
Sat Nov 25 11:06:14 2017 - [info] Locking all tables on the orig master to reject updates from everybody (including root):
Sat Nov 25 11:06:14 2017 - [info] Executing FLUSH TABLES WITH READ LOCK..
Sat Nov 25 11:06:14 2017 - [info]  ok.
Sat Nov 25 11:06:14 2017 - [info] Orig master binlog:pos is 11213389-bin.000003:194.
Sat Nov 25 11:06:14 2017 - [info]  Waiting to execute all relay logs on 192.168.1.120(192.168.1.120:3389)..
Sat Nov 25 11:06:14 2017 - [info]  master_pos_wait(11213389-bin.000003:194) completed on 192.168.1.120(192.168.1.120:3389). Executed 0 events.
Sat Nov 25 11:06:14 2017 - [info]   done.
Sat Nov 25 11:06:14 2017 - [info] Getting new master's binlog name and position..
Sat Nov 25 11:06:14 2017 - [info]  11203389-bin.000003:346
Sat Nov 25 11:06:14 2017 - [info]  All other slaves should start replication from here. Statement should be: CHANGE MASTER TO MASTER_HOST='192.168.1.120', MASTER_PORT=3389, MASTER_AUTO_POSITION=1, MASTER_USER='replica', MASTER_PASSWORD='xxx';
Sat Nov 25 11:06:14 2017 - [info] Executing master ip online change script to allow write on the new master:
Sat Nov 25 11:06:14 2017 - [info]   /etc/mha/master_ip_online_change.sh 192.168.1.200 1 --command=start --orig_master_host=192.168.1.121 --orig_master_ip=192.168.1.121 --orig_master_port=3389 --orig_master_user='mha' --new_master_host=192.168.1.120 --new_master_ip=192.168.1.120 --new_master_port=3389 --new_master_user='mha' --orig_master_ssh_user=mysql --new_master_ssh_user=mysql   --orig_master_is_new_slave --orig_master_password=xxx --new_master_password=xxx
Unknown option: orig_master_ssh_user
Unknown option: new_master_ssh_user
Unknown option: orig_master_is_new_slave
Sat Nov 25 11:06:14 2017 186596 Set read_only=0 on the new master.
Enabling the VIP - 192.168.1.200 on the new master - 192.168.1.120 
===========sudo /sbin/ifconfig eth1:1 192.168.1.200 broadcast 192.168.1.255 netmask 255.255.255.0 && sudo /sbin/arping -f -q -c 5 -w 5 -I eth1 -s 192.168.1.200  -U 192.168.1.1===========================
Sat Nov 25 11:06:14 2017 - [info]  ok.
Sat Nov 25 11:06:14 2017 - [info] 
Sat Nov 25 11:06:14 2017 - [info] * Switching slaves in parallel..
Sat Nov 25 11:06:14 2017 - [info] 
Sat Nov 25 11:06:14 2017 - [info] Unlocking all tables on the orig master:
Sat Nov 25 11:06:14 2017 - [info] Executing UNLOCK TABLES..
Sat Nov 25 11:06:14 2017 - [info]  ok.
Sat Nov 25 11:06:14 2017 - [info] Starting orig master as a new slave..
Sat Nov 25 11:06:14 2017 - [info]  Resetting slave 192.168.1.121(192.168.1.121:3389) and starting replication from the new master 192.168.1.120(192.168.1.120:3389)..
Sat Nov 25 11:06:14 2017 - [info]  Executed CHANGE MASTER.
Sat Nov 25 11:06:14 2017 - [info]  Slave started.
Sat Nov 25 11:06:14 2017 - [info] All new slave servers switched successfully.
Sat Nov 25 11:06:14 2017 - [info] 
Sat Nov 25 11:06:14 2017 - [info] * Phase 5: New master cleanup phase..
Sat Nov 25 11:06:14 2017 - [info] 
Sat Nov 25 11:06:14 2017 - [info]  192.168.1.120: Resetting slave info succeeded.
Sat Nov 25 11:06:14 2017 - [info] Switching master to 192.168.1.120(192.168.1.120:3389) completed successfully.

扫描下方二维码关注本人微信号!欢迎大家交流学习!

Bruce.Liu






推荐阅读更多精彩内容