MySQL 高可用架构 - MHA环境部署记录
一、MHA介绍
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,是日本的一位MySQL专家采用Perl语言编写的一个脚本管理工具,该工具仅适用于MySQLReplication(二层)环境,目的在于维持Master主库的高可用性。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能在最大程度上保证数据的一致性,以达到真正意义上的高可用。
MHA是自动的master故障转移和Slave提升的软件包.它是基于标准的MySQL复制(异步/半同步).该软件由两部分组成:MHA Manager(管理节点)和MHA Node(数据节点)。
1. 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 Manager会定时探测集群中的node节点,当发现master出现故障的时候,它可以自动将具有最新数据的slave提升为新的master,然后将所有其它的slave导向新的master上.整个故障转移过程对应用程序是透明的。
2. MHA Node运行在每台MySQL服务器上,它通过监控具备解析和清理logs功能的脚本来加快故障转移的。
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,最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最新的数据。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已经收到了最新的二进制日志,MHA可以将最新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。
目前MHA主要支持一主多从的架构,要搭建MHA,要求一个复制集群中必须最少有三台数据库服务器,一主二从,即一台充当master,一台充当备用master,另外一台充当从库,因为至少需要三台服务器,出于机器成本的考虑,淘宝也在该基础上进行了改造,目前淘宝TMHA已经支持一主一从。
二、MHA工作架构说明
展示了如何通过MHA Manager管理多组主从复制。可以将MHA工作原理总结为如下:
相较于其它HA软件,MHA的目的在于维持MySQL Replication中Master库的高可用性,其最大特点是可以修复多个Slave之间的差异日志,最终使所有Slave保持数据一致,然后从中选择一个充当新的Master,并将其它Slave指向它。工作流程主要如下:
1. 从宕机崩溃的master保存二进制日志事件(binlog events);
2. 识别含有最新更新的slave;
3. 应用差异的中继日志(relay log)到其他的slave;
4. 应用从master保存的二进制日志事件(binlog events);
5. 提升一个slave为新的master;
6. 使其他的slave连接新的master进行复制;
################ MHA工作原理 ###############
当master出现故障时,通过对比slave之间I/O线程读取master binlog的位置,选取最接近的slave做为latest slave。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。
############ MHA软件的架构 ########### 由两部分组成,Manager工具包 和 Node工具包,具体的说明如下。
1. Manager工具包主要包括以下几个工具:
masterha_check_ssh # 检查MHA的SSH配置状况
masterha_check_repl # 检查MySQL复制状况
masterha_manger # 启动MHA
masterha_check_status # 检测当前MHA运行状态
masterha_master_monitor # 检测master是否宕机
masterha_master_switch # 控制故障转移(自动或者手动)
masterha_conf_host # 添加或删除配置的server信息
2. Node工具包(这些工具通常由MHA Manager的脚本触发,无需人为操作)主要包括以下几个工具:
save_binary_logs #(保存二进制日志) 保存和复制master的二进制日志
apply_diff_relay_logs # (应用差异中继日志) 识别差异的中继日志事件并将其差异的事件应用于其他的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 # (清理中继日志) 清除中继日志(不会阻塞SQL线程)
########## MHA如何保持数据的一致性呢?######### 主要通过MHA node的以下几个工具实现,但是这些工具由mha manager触发:
save_binary_logs 如果master的二进制日志可以存取的话,保存复制master的二进制日志,最大程度保证数据不丢失
apply_diff_relay_logs 相对于最新的slave,生成差异的中继日志并将所有差异事件应用到其他所有的slave
注意:
对比的是relay log,relay log越新就越接近于master,才能保证数据是最新的。
purge_relay_logs删除中继日志而不阻塞sql线程
################# MHA的优势 ##################
1. 故障切换快
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。
2. master故障不会导致数据不一致
当目前的master出现故障时,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。
3. 无需修改当前的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。
4. 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。
5. 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。
6. 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。
三、MHA高可用环境部署记录
1. 机器环境
ip地址 主机名 角色 182.48.115.236 Node_Master 写入,数据节点 182.48.115.237 Node_Slave 读,数据节点,备选Master(candicate master) 182.48.115.238 Manager_Slave 读,数据节点,也作为Manager server(即也作为manager节点) ........................................................................................................ 为了节省机器,这里选择只读的从库182.48.115.237(从库不对外提供读的服务)作为候选主库,即candicate master,或是专门用于备份 同样,为了节省机器,这里选择182.48.115.238这台从库作为manager server(实际生产环节中,机器充足的情况下, 一般是专门选择一台机器作为Manager server) ........................................................................................................ 关闭三台机器的iptables和selinux 部署节点之间ssh无密码登陆的信任关系(即在所有节点上做ssh免密码登录,包括对节点本机的信任) [root@Node_Master ~]# ssh-copy-id 182.48.115.236 [root@Node_Master ~]# ssh-copy-id 182.48.115.237 [root@Node_Master ~]# ssh-copy-id 182.48.115.238 [root@Node_Slave ~]# ssh-copy-id 182.48.115.236 [root@Node_Slave ~]# ssh-copy-id 182.48.115.237 [root@Node_Slave ~]# ssh-copy-id 182.48.115.238 [root@Manager_Slave ~]# ssh-copy-id 182.48.115.236 [root@Manager_Slave ~]# ssh-copy-id 182.48.115.237 [root@Manager_Slave ~]# ssh-copy-id 182.48.115.238 现在3台节点已经能实现两两互相ssh通了,不需要输入密码即可。如果不能实现任何两台主机互相之间可以无密码登录,后面的环节可能会有问题。
2. 实现主机名hostname登录(在三台节点上都需要执行)(这一步不是必须要操作的)
分别设置三台节点机器的主机名(主机名上面已提出),并绑定hosts. 三台机器的/etc/hosts文件的绑定信息如下: [root@Node_Master ~]# vim /etc/hosts ....... 182.48.115.236 Node_Master 182.48.115.237 Node_Slave 182.48.115.238 Manager_Slave 相互验证下使用主机名登陆是否正常,是否可以相互使用主机名ssh无密码登陆到对方。
3. 准备好Mysql主从环境
架构如下,一主二从的架构: 主库:182.48.115.236 从库:182.48.115.237 主库:182.48.115.236 从库:182.48.115.238 Mysql主从环境部署可以参考:http://www.cnblogs.com/kevingrace/p/6256603.html ....................................................................................... ------主库配置------ server-id=1 log-bin=mysql-bin binlog-ignore-db=mysql sync_binlog = 1 binlog_checksum = none binlog_format = mixed ------从库1配置------- server-id=2 log-bin=mysql-bin binlog-ignore-db=mysql # 千万要注意:主从同步中的过滤字段要一致,否则后面使用masterha_check_repl 检查复制时就会出错! slave-skip-errors = all ------从库2配置------- server-id=3 log-bin=mysql-bin binlog-ignore-db=mysql slave-skip-errors = all 然后主库授权给从库连接的权限,设置后,最好在从库上验证下是否能使用授予的权限连接主库。 然后在从库上根据主库的“show master status;” 信心进行change master.....同步设置。 注意: 主从设置时,如果设置了bbinlog-ignore-db 和 replicate-ignore-db 过滤规则,则主从必须相同。即要使用binlog-ignore-db过滤字段,则主从配置都使用这个, 要是使用replicate-ignore-db过滤字段,则主从配置都使用这个,千万不能主从配置使用的过滤字段不一样!因为MHA 在启动时候会检测过滤规则,如果过滤规则不同,MHA 不启动监控和故障转移。 .......................................................................................
4. 创建用户mha管理的账号(在三台节点上都需要执行)
mysql> GRANT SUPER,RELOAD,REPLICATION CLIENT,SELECT ON *.* TO manager@'182.48.115.%' IDENTIFIED BY 'manager_1234'; Query OK, 0 rows affected (0.06 sec) mysql> GRANT CREATE,INSERT,UPDATE,DELETE,DROP ON*.* TO manager@'182.48.115.%'; Query OK, 0 rows affected (0.05 sec) 创建主从账号(在三台节点上都需要执行): mysql> GRANT RELOAD, SUPER, REPLICATION SLAVE ON*.* TO 'repl'@'182.48.115.%' IDENTIFIED BY 'repl_1234'; Query OK, 0 rows affected (0.09 sec) mysql> flush privileges; Query OK, 0 rows affected (0.06 sec)
5. 开始安装mha
mha包括manager节点和data节点,其中:
data节点包括原有的MySQL复制结构中的主机,至少3台,即1主2从,当master failover后,还能保证主从结构;只需安装node包。
manager server:运行监控脚本,负责monitoring 和 auto-failover;需要安装node包和manager包。
5.1 在所有data数据节点机上安装安装MHA node
下载mha4mysql-node-0.56.tar.gz 下载地址:http://pan.baidu.com/s/1cphgLo 提取密码:7674 [root@Node_Master ~]# yum -y install perl-DBD-MySQL //先安装所需的perl模块 [root@Node_Master ~]# tar -zvxf mha4mysql-node-0.56.tar.gz [root@Node_Master ~]# cd mha4mysql-node-0.56 [root@Node_Master mha4mysql-node-0.56]# perl Makefile.PL ................................................................................................................ 这一步可能报错如下: 1)Can't locate ExtUtils/MakeMaker.pm in @INC (@INC contains: inc /usr/local/lib64/perl5 /usr/local/share/perl5...... 解决办法: [root@Node_Master mha4mysql-node-0.56]# yum install perl-ExtUtils-CBuilder perl-ExtUtils-MakeMaker 2)Can't locate CPAN.pm in @INC (@INC contains: inc /usr/local/lib64/perl5 /usr/local/share/perl5 /usr/lib64/perl5.... 解决办法: [root@Node_Master mha4mysql-node-0.56]# yum install -y perl-CPAN ................................................................................................................ [root@Node_Master mha4mysql-node-0.56]# make && make install
5.2 在manager节点(即182.48.115.238)上安装MHA Manager(注意manager节点也要安装MHA node)
首先下载第三方yum源 [root@Manager_Slave ~]# rpm -ivh http://dl.fedoraproject.org/pub/epel/6/x86_64/epel-release-6-8.noarch.rpm 安装perl的mysql包: [root@Manager_Slave ~]# yum install -y perl-DBD-MySQL perl-Config-Tiny perl-Log-Dispatch perl-Parallel-ForkManager perl-Config-IniFiles perl-Time-HiRes -y 安装MHA Manager软件包: 下载地址:https://pan.baidu.com/s/1slyfXN3 提取密码:86wb [root@Manager_Slave ~]# tar -vxf mha4mysql-manager-0.56.tar [root@Manager_Slave ~]# cd mha4mysql-manager-0.56 [root@Manager_Slave mha4mysql-manager-0.56]# perl Makefile.PL [root@Manager_Slave mha4mysql-manager-0.56]# make && make install 安装完MHA Manager后,在/usr/local/bin目录下会生成以下脚本: [root@Manager_Slave mha4mysql-manager-0.56]# ll /usr/local/bin/ 总用量 84 -r-xr-xr-x. 1 root root 16367 5月 31 21:37 apply_diff_relay_logs -r-xr-xr-x. 1 root root 4807 5月 31 21:37 filter_mysqlbinlog -r-xr-xr-x. 1 root root 1995 5月 31 22:23 masterha_check_repl -r-xr-xr-x. 1 root root 1779 5月 31 22:23 masterha_check_ssh -r-xr-xr-x. 1 root root 1865 5月 31 22:23 masterha_check_status -r-xr-xr-x. 1 root root 3201 5月 31 22:23 masterha_conf_host -r-xr-xr-x. 1 root root 2517 5月 31 22:23 masterha_manager -r-xr-xr-x. 1 root root 2165 5月 31 22:23 masterha_master_monitor -r-xr-xr-x. 1 root root 2373 5月 31 22:23 masterha_master_switch -r-xr-xr-x. 1 root root 5171 5月 31 22:23 masterha_secondary_check -r-xr-xr-x. 1 root root 1739 5月 31 22:23 masterha_stop -r-xr-xr-x. 1 root root 8261 5月 31 21:37 purge_relay_logs -r-xr-xr-x. 1 root root 7525 5月 31 21:37 save_binary_logs 其中: masterha_check_repl 检查MySQL复制状况 masterha_check_ssh 检查MHA的SSH配置状况 masterha_check_status 检测当前MHA运行状态 masterha_conf_host 添加或删除配置的server信息 masterha_manager 启动MHA masterha_stop 停止MHA masterha_master_monitor 检测master是否宕机 masterha_master_switch 控制故障转移(自动或者手动) masterha_secondary_check 多种线路检测master是否存活 另外: 在../mha4mysql-manager-0.56/samples/scripts下还有以下脚本,需要将其复制到/usr/local/bin [root@Manager_Slave mha4mysql-manager-0.56]# cd samples/scripts/ [root@Manager_Slave scripts]# ll 总用量 32 -rwxr-xr-x. 1 4984 users 3648 4月 1 2014 master_ip_failover //自动切换时VIP管理脚本,不是必须,如果我们使用keepalived的,我们可以自己编写脚本完成对vip的管理,比如监控mysql,如果mysql异常,我们停止keepalived就行,这样vip就会自动漂移 -rwxr-xr-x. 1 4984 users 9870 4月 1 2014 master_ip_online_change //在线切换时VIP脚本,不是必须,同样可以可以自行编写简单的shell完成 -rwxr-xr-x. 1 4984 users 11867 4月 1 2014 power_manager //故障发生后关闭master脚本,不是必须 -rwxr-xr-x. 1 4984 users 1360 4月 1 2014 send_report //故障切换发送报警脚本,不是必须,可自行编写简单的shell完成 [root@Manager_Slave scripts]# cp ./* /usr/local/bin/
5.3 在管理节点(182.48.115.238)上进行下面配置
[root@Manager_Slave mha4mysql-manager-0.56]# mkdir -p /etc/masterha [root@Manager_Slave mha4mysql-manager-0.56]# cp samples/conf/app1.cnf /etc/masterha/ [root@Manager_Slave mha4mysql-manager-0.56]# vim /etc/masterha/app1.cnf [server default] manager_workdir=/var/log/masterha/app1 //设置manager的工作目录 manager_log=/var/log/masterha/app1/manager.log //设置manager的日志 ssh_user=root //ssh免密钥登录的帐号名 repl_user=repl //mysql复制帐号,用来在主从机之间同步二进制日志等 repl_password=repl_1234 //设置mysql中root用户的密码,这个密码是前文中创建监控用户的那个密码 ping_interval=1 //设置监控主库,发送ping包的时间间隔,用来检查master是否正常,默认是3秒,尝试三次没有回应的时候自动进行railover master_ip_failover_script= /usr/local/bin/master_ip_failover //设置自动failover时候的切换脚本 master_ip_online_change_script= /usr/local/bin/master_ip_online_change //设置手动切换时候的切换脚本 [server1] hostname=182.48.115.236 port=3306 master_binlog_dir=/data/mysql/data/ //设置master 保存binlog的位置,以便MHA可以找到master的日志,我这里的也就是mysql的数据目录 [server2] hostname=182.48.115.237 port=3306 candidate_master=1 //设置为候选master,即master机宕掉后,优先启用这台作为新master,如果设置该参数以后,发生主从切换以后将会将此从库提升为主库,即使这个主库不是集群中事件最新的slave check_repl_delay=0 //默认情况下如果一个slave落后master 100M的relay logs的话,MHA将不会选择该slave作为一个新的master,因为对于这个slave的恢复需要花费很长时间,通过设置check_repl_delay=0,MHA触发切换在选择一个新的master的时候将会忽略复制延时,这个参数对于设置了candidate_master=1的主机非常有用,因为这个候选主在切换的过程中一定是新的master master_binlog_dir=/data/mysql/data/ [server3] hostname=182.48.115.238 port=3306 #candidate_master=1 master_binlog_dir=/data/mysql/data/ #[server4] #hostname=host4 #no_master=1
5.4 设置relay log的清除方式(在两台slave节点上)
[root@Node_Slave ~]# mysql -p123456 -e 'set global relay_log_purge=0' [root@Manager_Slave ~]# mysql -p123456 -e 'set global relay_log_purge=0' .................................................................................................. 温馨提示: MHA在发生切换的过程中,从库的恢复过程中依赖于relay log的相关信息,所以这里要将relay log的自动清除设置为OFF,采用手动清除relay log的方式。 在默认情况下,从服务器上的中继日志会在SQL线程执行完毕后被自动删除。但是在MHA环境中,这些中继日志在恢复其他从服务器时可能会被用到,因此需要禁用 中继日志的自动删除功能。定期清除中继日志需要考虑到复制延时的问题。在ext3的文件系统下,删除大的文件需要一定的时间,会导致严重的复制延时。为了避 免复制延时,需要暂时为中继日志创建硬链接,因为在linux系统中通过硬链接删除大文件速度会很快。(在mysql数据库中,删除大表时,通常也采用建立硬链接的方式) MHA节点中包含了pure_relay_logs命令工具,它可以为中继日志创建硬链接,执行SET GLOBAL relay_log_purge=1,等待几秒钟以便SQL线程切换到新的中继日志, 再执行SET GLOBAL relay_log_purge=0。 pure_relay_logs脚本参数如下所示: --user mysql 用户名 --password mysql 密码 --port 端口号 --workdir 指定创建relay log的硬链接的位置,默认是/var/tmp,由于系统不同分区创建硬链接文件会失败,故需要执行硬链接具体位置,成功执行脚本后,硬链接的中继日志文件被删除 --disable_relay_log_purge 默认情况下,如果relay_log_purge=1,脚本会什么都不清理,自动退出,通过设定这个参数,当relay_log_purge=1的情况下会将relay_log_purge设置为0。清理relay log之后,最后将参数设置为OFF。 设置定期清理relay脚本(在两台slave节点上操作) [root@Node_Slave ~]# vim /root/purge_relay_log.sh #!/bin/bash user=root passwd=123456 port=3306 host=localhost log_dir='/data/masterha/log' work_dir='/data' purge='/usr/local/bin/purge_relay_logs' if [ ! -d $log_dir ] then mkdir $log_dir -p fi $purge --user=$user --host=$host --password=$passwd --disable_relay_log_purge --port=$port --workdir=$work_dir >> $log_dir/purge_relay_logs.log 2>&1 [root@Node_Slave ~]# chmod 755 /root/purge_relay_log.sh 添加到crontab定期执行 [root@Node_Slave ~]# crontab -e 0 4 * * * /bin/bash /root/purge_relay_log.sh purge_relay_logs脚本删除中继日志不会阻塞SQL线程。下面手动执行看看什么情况。 [root@Node_Slave ~]# /usr/local/bin/purge_relay_logs --user=root --host=localhost --password=123456 --disable_relay_log_purge --port=3306 --workdir=/data 2017-05-31 23:27:13: purge_relay_logs script started. Found relay_log.info: /data/mysql/data/relay-log.info Opening /data/mysql/data/mysql-relay-bin.000002 .. Opening /data/mysql/data/mysql-relay-bin.000003 .. Executing SET GLOBAL relay_log_purge=1; FLUSH LOGS; sleeping a few seconds so that SQL thread can delete older relay log files (if it keeps up); SET GLOBAL relay_log_purge=0; .. ok. 2017-05-31 23:27:17: All relay log purging operations succeeded. [root@Node_Slave ~]# ll /data/masterha/log/ 总用量 4 -rw-r--r--. 1 root root 905 5月 31 23:26 purge_relay_logs.log
5.5 检查SSH配置
检查MHA Manger到所有MHA Node的SSH连接状态: [root@Manager_Slave ~]# masterha_check_ssh --conf=/etc/masterha/app1.cnf Wed May 31 23:06:01 2017 - [warning] Global configuration file /etc/masterha_default.cnf not found. Skipping. Wed May 31 23:06:01 2017 - [info] Reading application default configuration from /etc/masterha/app1.cnf.. Wed May 31 23:06:01 2017 - [info] Reading server configuration from /etc/masterha/app1.cnf.. Wed May 31 23:06:01 2017 - [info] Starting SSH connection tests.. Wed May 31 23:06:04 2017 - [debug] Wed May 31 23:06:01 2017 - [debug] Connecting via SSH from root@182.48.115.236(182.48.115.236:22) to root@182.48.115.237(182.48.115.237:22).. Wed May 31 23:06:02 2017 - [debug] ok. Wed May 31 23:06:02 2017 - [debug] Connecting via SSH from root@182.48.115.236(182.48.115.236:22) to root@182.48.115.238(182.48.115.238:22).. Wed May 31 23:06:03 2017 - [debug] ok. Wed May 31 23:06:04 2017 - [debug] Wed May 31 23:06:01 2017 - [debug] Connecting via SSH from root@182.48.115.237(182.48.115.237:22) to root@182.48.115.236(182.48.115.236:22).. Wed May 31 23:06:03 2017 - [debug] ok. Wed May 31 23:06:03 2017 - [debug] Connecting via SSH from root@182.48.115.237(182.48.115.237:22) to root@182.48.115.238(182.48.115.238:22).. Wed May 31 23:06:04 2017 - [debug] ok. Wed May 31 23:06:04 2017 - [debug] Wed May 31 23:06:02 2017 - [debug] Connecting via SSH from root@182.48.115.238(182.48.115.238:22) to root@182.48.115.236(182.48.115.236:22).. Wed May 31 23:06:03 2017 - [debug] ok. Wed May 31 23:06:03 2017 - [debug] Connecting via SSH from root@182.48.115.238(182.48.115.238:22) to root@182.48.115.237(182.48.115.237:22).. Wed May 31 23:06:04 2017 - [debug] ok. Wed May 31 23:06:04 2017 - [info] All SSH connection tests passed successfully. 可以看见各个节点ssh验证都是ok的。
5.6 使用mha工具check检查repl环境
通过masterha_check_repl脚本查看整个mysql集群的复制状态 [root@Manager_Slave ~]# masterha_check_repl --conf=/etc/masterha/app1.cnf Wed May 31 23:43:43 2017 - [warning] Global configuration file /etc/masterha_default.cnf not found. Skipping. Wed May 31 23:43:43 2017 - [info] Reading application default configuration from /etc/masterha/app1.cnf.. Wed May 31 23:43:43 2017 - [info] Reading server configuration from /etc/masterha/app1.cnf.. Wed May 31 23:43:43 2017 - [info] MHA::MasterMonitor version 0.56. Wed May 31 23:43:43 2017 - [error][/usr/local/share/perl5/MHA/ServerManager.pm, ln301] Got MySQL error when connecting 182.48.115.237(182.48.115.237:3306) :1045:Access denied for user 'root'@'182.48.115.238' (using password: NO), but this is not a MySQL crash. Check MySQL server settings. at /usr/local/share/perl5/MHA/ServerManager.pm line 297 Wed May 31 23:43:43 2017 - [error][/usr/local/share/perl5/MHA/ServerManager.pm, ln301] Got MySQL error when connecting 182.48.115.236(182.48.115.236:3306) :1045:Access denied for user 'root'@'182.48.115.238' (using password: NO), but this is not a MySQL crash. Check MySQL server settings. at /usr/local/share/perl5/MHA/ServerManager.pm line 297 Wed May 31 23:43:43 2017 - [error][/usr/local/share/perl5/MHA/ServerManager.pm, ln301] Got MySQL error when connecting 182.48.115.238(182.48.115.238:3306) :1045:Access denied for user 'root'@'182.48.115.238' (using password: NO), but this is not a MySQL crash. Check MySQL server settings. at /usr/local/share/perl5/MHA/ServerManager.pm line 297 Wed May 31 23:43:43 2017 - [error][/usr/local/share/perl5/MHA/ServerManager.pm, ln309] Got fatal error, stopping operations Wed May 31 23:43:43 2017 - [error][/usr/local/share/perl5/MHA/MasterMonitor.pm, ln424] Error happened on checking configurations. at /usr/local/share/perl5/MHA/MasterMonitor.pm line 326 Wed May 31 23:43:43 2017 - [error][/usr/local/share/perl5/MHA/MasterMonitor.pm, ln523] Error happened on monitoring servers. Wed May 31 23:43:43 2017 - [info] Got exit code 1 (Not master dead). MySQL Replication Health is NOT OK! 发现上面的复制环节是不ok的!!! 原因是通过root用户远程连接节点的mysql不通 .............................................................................................................. 解决办法:在三个节点机器上的mysql上授权,允许182.48.115.%的机器通过root用户无密码登陆,即 mysql> update mysql.user set password=password("") where user="root" and host="182.48.115.%"; //如果没有这个权限,就grant命令创建这个用户权限 Query OK, 1 row affected (0.00 sec) Rows matched: 1 Changed: 1 Warnings: 0 mysql> flush privileges; Query OK, 0 rows affected (0.00 sec) mysql> select user,host,password from mysql.user; +---------+--------------+-------------------------------------------+ | user | host | password | +---------+--------------+-------------------------------------------+ ......... | root | 182.48.115.% | | +---------+--------------+-------------------------------------------+ 11 rows in set (0.00 sec) .............................................................................................................. 然后再次通过masterha_check_repl脚本查看整个mysql集群的复制状态 [root@Manager_Slave ~]# masterha_check_repl --conf=/etc/masterha/app1.cnf .............................. Bareword "FIXME_xxx" not allowed while "strict subs" in use at /usr/local/bin/master_ip_failover line 93. 还是出现如上报错,原因是: 原来Failover两种方式:一种是虚拟IP地址,一种是全局配置文件。MHA并没有限定使用哪一种方式,而是让用户自己选择,虚拟IP地址的方式会牵扯到其它的软件,比如keepalive软件,而且还要修改脚本master_ip_failover。 解决办法如下: 添加软连接(所有节点): [root@Manager_Slave ~]# ln -s /usr/local/mysql/bin/mysqlbinlog /usr/local/bin/mysqlbinlog [root@Manager_Slave ~]# ln -s /usr/local/mysql/bin/mysql /usr/local/bin/mysql 先暂时注释掉管理节点的/etc/masterha/app1.cnf文件中的master_ip_failover_script= /usr/local/bin/master_ip_failover这个选项。 后面引入keepalived后和修改该脚本以后再开启该选项。 [root@Manager_Slave ~]# cat /etc/masterha/app1.cnf ......... #master_ip_failover_script= /usr/local/bin/master_ip_failover 最后在通过masterha_check_repl脚本查看整个mysql集群的复制状态 [root@Manager_Slave ~]# masterha_check_repl --conf=/etc/masterha/app1.cnf Thu Jun 1 00:20:58 2017 - [warning] Global configuration file /etc/masterha_default.cnf not found. Skipping. Thu Jun 1 00:20:58 2017 - [info] Reading application default configuration from /etc/masterha/app1.cnf.. Thu Jun 1 00:20:58 2017 - [info] read_only=1 is not set on slave 182.48.115.237(182.48.115.237:3306). Thu Jun 1 00:20:58 2017 - [warning] relay_log_purge=0 is not set on slave 182.48.115.237(182.48.115.237:3306). Thu Jun 1 00:20:58 2017 - [info] read_only=1 is not set on slave 182.48.115.238(182.48.115.238:3306). Thu Jun 1 00:20:58 2017 - [warning] relay_log_purge=0 is not set on slave 182.48.115.238(182.48.115.238:3306). Thu Jun 1 00:20:58 2017 - [info] Checking replication filtering settings.. Thu Jun 1 00:20:58 2017 - [info] binlog_do_db= , binlog_ignore_db= mysql Thu Jun 1 00:20:58 2017 - [info] Replication filtering check ok. Thu Jun 1 00:20:58 2017 - [info] GTID (with auto-pos) is not supported Thu Jun 1 00:20:58 2017 - [info] Starting SSH connection tests.. Thu Jun 1 00:21:02 2017 - [info] All SSH connection tests passed successfully. ........... Thu Jun 1 00:21:07 2017 - [info] Checking replication health on 182.48.115.237.. Thu Jun 1 00:21:07 2017 - [info] ok. Thu Jun 1 00:21:07 2017 - [info] Checking replication health on 182.48.115.238.. Thu Jun 1 00:21:07 2017 - [info] ok. Thu Jun 1 00:21:07 2017 - [warning] master_ip_failover_script is not defined. Thu Jun 1 00:21:07 2017 - [warning] shutdown_script is not defined. Thu Jun 1 00:21:07 2017 - [info] Got exit code 0 (Not master dead). MySQL Replication Health is OK. 这个时候,发现整个复制环境状况是ok的了!!
6. 管理mha操作
6.1 检查MHA Manager的状态
通过master_check_status脚本查看Manager的状态 [root@Manager_Slave ~]# masterha_check_status --conf=/etc/masterha/app1.cnf app1 is stopped(2:NOT_RUNNING). 注意:如果正常,会显示"PING_OK",否则会显示"NOT_RUNNING",这代表MHA监控没有开启
6.2 开启MHA Manager监控
使用下面命令放在后台执行启动动作 [root@Manager_Slave ~]# nohup masterha_manager --conf=/etc/masterha/app1.cnf --remove_dead_master_conf --ignore_last_failover < /dev/null > /var/log/masterha/app1/manager.log 2>&1 & 启动参数介绍: --remove_dead_master_conf 该参数代表当发生主从切换后,老的主库的ip将会从配置文件中移除。 --manger_log 日志存放位置 --ignore_last_failover 在缺省情况下,如果MHA检测到连续发生宕机,且两次宕机间隔不足8小时的话,则不会进行Failover,之所以这样限制是为了 避免ping-pong效应。该参数代表忽略上次MHA触发切换产生的文件,默认情况下,MHA发生切换后会在日志目录,也就是上面我 设置的/data产生app1.failover.complete文件,下次再次切换的时候如果发现该目录下存在该文件将不允许触发切换,除非 在第一次切换后收到删除该文件,为了方便,这里设置为--ignore_last_failover。 再次查看MHA Manager监控是否正常: [root@Manager_Slave ~]# masterha_check_status --conf=/etc/masterha/app1.cnf app1 (pid:2542) is running(0:PING_OK), master:182.48.115.236 可以看见已经在监控了,而且master的主机为182.48.115.236 查看启动日志 [root@Manager_Slave ~]# tail -n20 /var/log/masterha/app1/manager.log Checking slave recovery environment settings.. Opening /data/mysql/data/relay-log.info ... ok. Relay log found at /data/mysql/data, up to mysql-relay-bin.000006 Temporary relay log file is /data/mysql/data/mysql-relay-bin.000006 Testing mysql connection and privileges..Warning: Using a password on the command line interface can be insecure. done. Testing mysqlbinlog output.. done. Cleaning up test file(s).. done. Thu Jun 1 00:37:29 2017 - [info] Slaves settings check done. Thu Jun 1 00:37:29 2017 - [info] 182.48.115.236(182.48.115.236:3306) (current master) +--182.48.115.237(182.48.115.237:3306) +--182.48.115.238(182.48.115.238:3306) Thu Jun 1 00:37:29 2017 - [warning] master_ip_failover_script is not defined. Thu Jun 1 00:37:29 2017 - [warning] shutdown_script is not defined. Thu Jun 1 00:37:29 2017 - [info] Set master ping interval 1 seconds. Thu Jun 1 00:37:29 2017 - [warning] secondary_check_script is not defined. It is highly recommended setting it to check master reachability from two or more routes. Thu Jun 1 00:37:29 2017 - [info] Starting ping health check on 182.48.115.236(182.48.115.236:3306).. Thu Jun 1 00:37:29 2017 - [info] Ping(SELECT) succeeded, waiting until MySQL doesn't respond.. 其中"Ping(SELECT) succeeded, waiting until MySQL doesn't respond.."说明整个系统已经开始监控了。
6.3 关闭MHA Manage监控
关闭很简单,使用masterha_stop命令完成。 [root@Manager_Slave ~]# masterha_stop --conf=/etc/masterha/app1.cnf Stopped app1 successfully. [1]+ Exit 1 nohup masterha_manager --conf=/etc/masterha/app1.cnf --remove_dead_master_conf --ignore_last_failover < /dev/null > /var/log/masterha/app1/manager.log 2>&1 [root@Manager_Slave ~]# 查看MHA Manager监控,发现已关闭 [root@Manager_Slave ~]# masterha_check_status --conf=/etc/masterha/app1.cnf app1 is stopped(2:NOT_RUNNING).
7. 配置VIP
vip配置可以采用两种方式,一种通过keepalived的方式管理虚拟ip浮动;另外一种通过脚本方式启动虚拟ip的方式(即不需要keepalived或者heartbeat类似的软件)。
第一种方式:通过keepalive的方式管理vip
1. 下载软件进行并进行安装(在两台master上都要安装,准确的说一台是master(182.48.115.236);另外一台是备选master(182.48.115.237),在没有切换以前是slave)
[root@Node_Master ~]# yum install -y openssl-devel [root@Node_Master ~]# wget http://www.keepalived.org/software/keepalived-1.3.5.tar.gz [root@Node_Master ~]# tar -zvxf keepalived-1.3.5.tar.gz [root@Node_Master ~]# cd keepalived-1.3.5 [root@Node_Master keepalived-1.3.5]# ./configure --prefix=/usr/local/keepalived [root@Node_Master keepalived-1.3.5]# make && make install [root@Node_Master keepalived-1.3.5]# cp keepalived/etc/init.d/keepalived /etc/init.d/ [root@Node_Master keepalived-1.3.5]# cp /usr/local/keepalived/etc/sysconfig/keepalived /etc/sysconfig/ [root@Node_Master keepalived-1.3.5]# mkdir /etc/keepalived [root@Node_Master keepalived-1.3.5]# cp /usr/local/keepalived/etc/keepalived/keepalived.conf /etc/keepalived/ [root@Node_Master keepalived-1.3.5]# cp /usr/local/keepalived/sbin/keepalived /usr/sbin/
2. keepalived配置
------------在master上配置(182.48.115.236节点上的配置)------------------ [root@Node_Master ~]# cp /etc/keepalived/keepalived.conf /etc/keepalived/keepalived.conf.bak [root@Node_Master ~]# vim /etc/keepalived/keepalived.conf ! Configuration File for keepalived global_defs { notification_email { wangshibo@huanqiu.cn } notification_email_from ops@huanqiu.cn smtp_server 127.0.0.1 smtp_connect_timeout 30 router_id MySQL-HA } vrrp_instance VI_1 { state BACKUP interface eth1 virtual_router_id 51 priority 150 advert_int 1 nopreempt authentication { auth_type PASS auth_pass 1111 } virtual_ipaddress { 182.48.115.239 } } 其中router_id MySQL HA表示设定keepalived组的名称,将182.48.115.239这个虚拟ip绑定到该主机的eth1网卡上,并且设置了状态为backup模式, 将keepalived的模式设置为非抢占模式(nopreempt),priority 150表示设置的优先级为150。 ------------在candicate master上配置(182.48.115.237节点上的配置)------------------ [root@Node_Slave ~]# vim /etc/keepalived/keepalived.conf ! Configuration File for keepalived global_defs { notification_email { wangshibo@huanqiu.cn } notification_email_from ops@huanqiu.cn smtp_server 127.0.0.1 smtp_connect_timeout 30 router_id MySQL-HA } vrrp_instance VI_1 { state BACKUP interface eth1 virtual_router_id 51 priority 120 advert_int 1 nopreempt authentication { auth_type PASS auth_pass 1111 } virtual_ipaddress { 182.48.115.239 } }
3. 启动keepalived服务
--------------在master上启动并查看日志---------------- [root@Node_Master ~]# /etc/init.d/keepalived start 正在启动 keepalived: [确定] [root@Node_Master ~]# ip addr 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo inet6 ::1/128 scope host valid_lft forever preferred_lft forever 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000 link/ether 52:54:00:5f:58:dc brd ff:ff:ff:ff:ff:ff inet 182.48.115.236/27 brd 182.48.115.255 scope global eth0 inet 182.48.115.239/32 scope global eth0 inet6 fe80::5054:ff:fe5f:58dc/64 scope link valid_lft forever preferred_lft forever [root@Node_Master ~]# tail -100 /var/log/messages .......... Jun 1 02:12:10 percona1 Keepalived_vrrp[10329]: VRRP_Instance(VI_1) Sending/queueing gratuitous ARPs on eth0 for 182.48.115.239 Jun 1 02:12:10 percona1 Keepalived_vrrp[10329]: Sending gratuitous ARP on eth0 for 182.48.115.239 Jun 1 02:12:10 percona1 Keepalived_vrrp[10329]: Sending gratuitous ARP on eth0 for 182.48.115.239 Jun 1 02:12:10 percona1 Keepalived_vrrp[10329]: Sending gratuitous ARP on eth0 for 182.48.115.239 Jun 1 02:12:10 percona1 Keepalived_vrrp[10329]: Sending gratuitous ARP on eth0 for 182.48.115.239 发现vip资源已经绑定到182.48.115.236这个master节点机上了 --------------在candicate master上启动---------------- [root@Node_Slave ~]# /etc/init.d/keepalived start 正在启动 keepalived: [确定] [root@Node_Slave ~]# ip addr 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo inet6 ::1/128 scope host valid_lft forever preferred_lft forever 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000 link/ether 52:54:00:1b:6e:53 brd ff:ff:ff:ff:ff:ff inet 182.48.115.237/27 brd 182.48.115.255 scope global eth0 inet6 fe80::5054:ff:fe1b:6e53/64 scope link valid_lft forever preferred_lft forever ..................................................................... 从上面的信息可以看到keepalived已经配置成功。 注意: 上面两台服务器的keepalived都设置为了BACKUP模式,在keepalived中2种模式,分别是master->backup模式和backup->backup模式。这两种模式有很大区别。 在master->backup模式下,一旦主库宕机,虚拟ip会自动漂移到从库,当主库修复后,keepalived启动后,还会把虚拟ip抢占过来,即使设置了非抢占模式(nopreempt) 抢占ip的动作也会发生。在backup->backup模式下,当主库宕机后虚拟ip会自动漂移到从库上,当原主库恢复和keepalived服务启动后,并不会抢占新主的虚拟ip,即使是 优先级高于从库的优先级别,也不会发生抢占。为了减少ip漂移次数,通常是把修复好的主库当做新的备库。
4. MHA引入keepalived(MySQL服务进程挂掉时通过MHA 停止keepalived)
要想把keepalived服务引入MHA,只需要修改切换是触发的脚本文件master_ip_failover即可,在该脚本中添加在master发生宕机时对keepalived的处理。 编辑脚本/usr/local/bin/master_ip_failover,修改后如下: [root@Manager_Slave ~]# vim /usr/local/bin/master_ip_failover #!/usr/bin/env perl use strict; use warnings FATAL => 'all'; use Getopt::Long; my ( $command, $ssh_user, $orig_master_host, $orig_master_ip, $orig_master_port, $new_master_host, $new_master_ip, $new_master_port ); my $vip = '182.48.115.239'; my $ssh_start_vip = "/etc/init.d/keepalived start"; my $ssh_stop_vip = "/etc/init.d/keepalived stop"; GetOptions( 'command=s' => \$command, 'ssh_user=s' => \$ssh_user, 'orig_master_host=s' => \$orig_master_host, 'orig_master_ip=s' => \$orig_master_ip, 'orig_master_port=i' => \$orig_master_port, 'new_master_host=s' => \$new_master_host, 'new_master_ip=s' => \$new_master_ip, 'new_master_port=i' => \$new_master_port, ); exit &main(); sub main { print "\n\nIN SCRIPT TEST====$ssh_stop_vip==$ssh_start_vip===\n\n"; if ( $command eq "stop" || $command eq "stopssh" ) { my $exit_code = 1; eval { print "Disabling the VIP on old master: $orig_master_host \n"; &stop_vip(); $exit_code = 0; }; if ($@) { warn "Got Error: $@\n"; exit $exit_code; } exit $exit_code; } elsif ( $command eq "start" ) { my $exit_code = 10; eval { print "Enabling the VIP - $vip on the new master - $new_master_host \n"; &start_vip(); $exit_code = 0; }; if ($@) { warn $@; exit $exit_code; } exit $exit_code; } elsif ( $command eq "status" ) { print "Checking the Status of the script.. OK \n"; #`ssh $ssh_user\@cluster1 \" $ssh_start_vip \"`; exit 0; } else { &usage(); exit 1; } } # A simple system call that enable the VIP on the new master sub start_vip() { `ssh $ssh_user\@$new_master_host \" $ssh_start_vip \"`; } # A simple system call that disable the VIP on the old_master sub stop_vip() { return 0 unless ($ssh_user); `ssh $ssh_user\@$orig_master_host \" $ssh_stop_vip \"`; } sub usage { print "Usage: master_ip_failover --command=start|stop|stopssh|status --orig_master_host=host --orig_master_ip=ip --orig_master_port=port --new_master_host=host --new_master_ip=ip --new_master_port=port\n"; } 现在已经修改这个脚本了,现在打开在上面提到过的参数,再检查集群状态,看是否会报错 [root@Manager_Slave ~]# grep 'master_ip_failover_script' /etc/masterha/app1.cnf master_ip_failover_script= /usr/local/bin/master_ip_failover [root@Manager_Slave ~]# masterha_check_repl --conf=/etc/masterha/app1.cnf ....... Checking the Status of the script.. OK Thu Jun 1 03:31:57 2017 - [info] OK. Thu Jun 1 03:31:57 2017 - [warning] shutdown_script is not defined. Thu Jun 1 03:31:57 2017 - [info] Got exit code 0 (Not master dead). MySQL Replication Health is OK. 可以看出复制情况正常! /usr/local/bin/master_ip_failover添加或者修改的内容意思是当主库数据库发生故障时,会触发MHA切换,MHA Manager会停掉主库上的keepalived服务, 触发虚拟ip漂移到备选从库,从而完成切换。当然可以在keepalived里面引入脚本,这个脚本监控mysql是否正常运行,如果不正常,则调用该脚本杀掉keepalived进程。
第二种方式:通过脚本的方式管理VIP
这里是修改/usr/local/bin/master_ip_failover,修改完成后内容如下。还需要手动在master服务器上绑定一个vip
1. 现在master节点上绑定vip
[root@Master_node ~]# ifconfig eth0:0 182.48.115.239/27 //本机子网掩码是27,一般都是24 [root@Master_node ~]# ifconfig eth0 Link encap:Ethernet HWaddr 52:54:00:5F:58:DC inet addr:182.48.115.236 Bcast:182.48.115.255 Mask:255.255.255.224 inet6 addr: fe80::5054:ff:fe5f:58dc/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:25505 errors:0 dropped:0 overruns:0 frame:0 TX packets:3358 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:1000 RX bytes:3254957 (3.1 MiB) TX bytes:482420 (471.1 KiB) eth0:0 Link encap:Ethernet HWaddr 52:54:00:5F:58:DC inet addr:182.48.115.239 Bcast:182.48.115.255 Mask:255.255.255.224 UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 lo Link encap:Local Loopback inet addr:127.0.0.1 Mask:255.0.0.0 inet6 addr: ::1/128 Scope:Host UP LOOPBACK RUNNING MTU:65536 Metric:1 RX packets:0 errors:0 dropped:0 overruns:0 frame:0 TX packets:0 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:0 RX bytes:0 (0.0 b) TX bytes:0 (0.0 b)
2. manager节点修改/usr/local/bin/master_ip_failover
[root@Manager_Slave ~]# cat /usr/local/bin/master_ip_failover #!/usr/bin/env perl use strict; use warnings FATAL => 'all'; use Getopt::Long; my ( $command, $ssh_user, $orig_master_host, $orig_master_ip, $orig_master_port, $new_master_host, $new_master_ip, $new_master_port ); my $vip = '182.48.115.239/27'; my $key = '1'; my $ssh_start_vip = "/sbin/ifconfig eth0:$key $vip"; my $ssh_stop_vip = "/sbin/ifconfig eth0:$key down"; GetOptions( 'command=s' => \$command, 'ssh_user=s' => \$ssh_user, 'orig_master_host=s' => \$orig_master_host, 'orig_master_ip=s' => \$orig_master_ip, 'orig_master_port=i' => \$orig_master_port, 'new_master_host=s' => \$new_master_host, 'new_master_ip=s' => \$new_master_ip, 'new_master_port=i' => \$new_master_port, ); exit &main(); sub main { print "\n\nIN SCRIPT TEST====$ssh_stop_vip==$ssh_start_vip===\n\n"; if ( $command eq "stop" || $command eq "stopssh" ) { my $exit_code = 1; eval { print "Disabling the VIP on old master: $orig_master_host \n"; &stop_vip(); $exit_code = 0; }; if ($@) { warn "Got Error: $@\n"; exit $exit_code; } exit $exit_code; } elsif ( $command eq "start" ) { my $exit_code = 10; eval { print "Enabling the VIP - $vip on the new master - $new_master_host \n"; &start_vip(); $exit_code = 0; }; if ($@) { warn $@; exit $exit_code; } exit $exit_code; } elsif ( $command eq "status" ) { print "Checking the Status of the script.. OK \n"; exit 0; } else { &usage(); exit 1; } } sub start_vip() { `ssh $ssh_user\@$new_master_host \" $ssh_start_vip \"`; } sub stop_vip() { return 0 unless ($ssh_user); `ssh $ssh_user\@$orig_master_host \" $ssh_stop_vip \"`; } sub usage { print "Usage: master_ip_failover --command=start|stop|stopssh|status --orig_master_host=host --orig_master_ip=ip --orig_master_port=port --new_master_host=host --new_master_ip=ip --new_master_port=port\n"; } 注意要将/etc/masterha/app1.cnf文件中的master_ip_failover_script注释打开 为了防止脑裂发生,推荐生产环境采用脚本的方式来管理虚拟ip,而不是使用keepalived来完成。到此为止,基本MHA集群已经配置完毕。 接下来就是实际的测试环节了。通过一些测试来看一下MHA到底是如何进行工作的。
8. failover故障切换
1. 自动切换(必须先启动MHA Manager,否则无法自动切换。(当然手动切换不需要开启MHA Manager监控))
1)在master主库上使用sysbench生成测试数据 [root@Master_node ~]# yum install sysbench -y 在主库(182.48.115.236)上进行sysbench数据生成,在sbtest库下生成sbtest表,共100W记录。 [root@Master_node ~]# sysbench --test=oltp --oltp-table-size=1000000 --oltp-read-only=off --init-rng=on --num-threads=16 --max-requests=0 --oltp-dist-type=uniform --max-time=1800 --mysql-user=root --mysql-socket=/local/mysql/var/mysql.sock --mysql-password=123456 --db-driver=mysql --mysql-table-engine=innodb --oltp-test-mode=complex prepare
1.1 在candicate master(182.48.115.237)上停掉slave sql线程,模拟主从延时。
mysql> stop slave io_thread; Query OK, 0 rows affected (0.08 sec) 注意:另外一台slave没有停止io线程,所以还在继续接收日志。
1.2 模拟sysbench压力测试
在主库上(182.48.115.236)进行压力测试,持续时间为3分钟,产生大量的binlog [root@Master_node ~]# sysbench --test=oltp --oltp-table-size=1000000 --oltp-read-only=off --init-rng=on --num-threads=16 --max-requests=0 --oltp-dist-type=uniform --max-time=180 --mysql-user=root --mysql-socket=/local/mysql/var/mysql.sock --mysql-password=123456 --db-driver=mysql --mysql-table-engine=innodb --oltp-test-mode=complex run
1.3 开启在candicate master(182.48.115.237)上的IO线程,追赶落后于master的binlog
mysql> start slave io_thread; Query OK, 0 rows affected (0.00 sec)
1.4 杀掉主库(182.48.115.236)mysql进程,模拟主库发生故障,进行自动failover操作
[root@Master_node ~]# pkill -9 mysqld
1.5 查看MHA切换日志,了解整个切换过程。在manager管理节点(182.48.115.238)上查看日志
[root@Manager_Slave ~]# cat /var/log/masterha/app1/manager.log ........ ........ ----- Failover Report ----- app1: MySQL Master failover 182.48.115.236 to 182.48.115.237 succeeded Master 182.48.115.236 is down! Check MHA Manager logs at server01:/var/log/masterha/app1/manager.log for details. Started automated(non-interactive) failover. Invalidated master IP address on 182.48.115.236. The latest slave 182.48.115.237(182.48.115.237:3306) has all relay logs for recovery. Selected 182.48.115.237 as a new master. 182.48.115.237: OK: Applying all logs succeeded. 182.48.115.237: OK: Activated master IP address. 192.168.0.70: This host has the latest relay log events. Generating relay diff files from the latest slave succeeded. 192.168.0.70: OK: Applying all logs succeeded. Slave started, replicating from 182.48.115.237. 182.48.115.237: Resetting slave info succeeded. Master failover to 182.48.115.237(182.48.115.237:3306) completed successfully. 看到最后的Master failover to 182.48.115.237(182.48.115.237:3306) completed successfully.说明备选master现在已经上位了。 从上面的输出可以看出整个MHA的切换过程,共包括以下的步骤: 1)配置文件检查阶段,这个阶段会检查整个集群配置文件配置 2)宕机的master处理,这个阶段包括虚拟ip摘除操作,主机关机操作(这个我这里还没有实现,需要研究) 3)复制dead maste和最新slave相差的relay log,并保存到MHA Manger具体的目录下 4)识别含有最新更新的slave 5)应用从master保存的二进制日志事件(binlog events) 6)提升一个slave为新的master进行复制 7)使其他的slave连接新的master进行复制 最后启动MHA Manger监控,查看集群里面现在谁是master [root@Manager_Slave ~]# masterha_check_status --conf=/etc/masterha/app1.cnf app1 (pid:13301) is running(0:PING_OK), master:182.48.115.237
2. 手动Failover(MHA Manager必须没有运行)
手动failover,这种场景意味着在业务上没有启用MHA自动切换功能,当主服务器故障时,人工手动调用MHA来进行故障切换操作,具体命令如下: 确保mha manager关闭 [root@Manager_Slave ~]# masterha_stop --conf=/etc/masterha/app1.cnf 注意:如果MHA manager检测到没有dead的server,将报错,并结束failover: [root@Manager_Slave ~]# masterha_master_switch --master_state=dead --conf=/etc/masterha/app1.cnf --dead_master_host=182.48.115.236 --dead_master_port=3306 --new_master_host=182.48.115.237 --new_master_port=3306 --ignore_last_failover 输出的信息会询问你是否进行切换: ........ ----- Failover Report ----- app1: MySQL Master failover 182.48.115.236 to 182.48.115.237 succeeded Master 182.48.115.236 is down! Check MHA Manager logs at server01 for details. Started manual(interactive) failover. Invalidated master IP address on 182.48.115.236. The latest slave 182.48.115.237(182.48.115.237:3306) has all relay logs for recovery. Selected 182.48.115.237 as a new master. 182.48.115.237: OK: Applying all logs succeeded. 182.48.115.237: OK: Activated master IP address. 192.168.0.70: This host has the latest relay log events. Generating relay diff files from the latest slave succeeded. 192.168.0.70: OK: Applying all logs succeeded. Slave started, replicating from 182.48.115.237. 182.48.115.237: Resetting slave info succeeded. Master failover to 182.48.115.237(182.48.115.237:3306) completed successfully. 这样即模拟了master宕机的情况下手动把192.168.0.60提升为主库的操作过程。
9. 在线进行切换
在许多情况下, 需要将现有的主服务器迁移到另外一台服务器上,比如主服务器硬件故障,RAID 控制卡需要重建,将主服务器移到性能更好的服务器上等等。维护主服务器引起性能下降, 导致停机时间至少无法写入数据。 另外, 阻塞或杀掉当前运行的会话会导致主主之间数据不一致的问题发生。 MHA提供快速切换和优雅的阻塞写入,这个切换过程只需要 0.5-2s 的时 间,这段时间内数据是无法写入的。在很多情况下,0.5-2s 的阻塞写入是可以接受的。因此切换主服务器不需要计划分配维护时间窗口。 MHA在线切换的大概过程: 1)检测复制设置和确定当前主服务器 2)确定新的主服务器 3)阻塞写入到当前主服务器 4)等待所有从服务器赶上复制 5)授予写入到新的主服务器 6)重新设置从服务器 注意,在线切换的时候应用架构需要考虑以下两个问题: 1)自动识别master和slave的问题(master的机器可能会切换),如果采用了vip的方式,基本可以解决这个问题。 2)负载均衡的问题(可以定义大概的读写比例,每台机器可承担的负载比例,当有机器离开集群时,需要考虑这个问题) 为了保证数据完全一致性,在最快的时间内完成切换,MHA的在线切换必须满足以下条件才会切换成功,否则会切换失败。 1)所有slave的IO线程都在运行 2)所有slave的SQL线程都在运行 3)所有的show slave status的输出中Seconds_Behind_Master参数小于或者等于running_updates_limit秒,如果在切换过程中不指定running_updates_limit,那么 默认情况下running_updates_limit为1秒。 4)在master端,通过show processlist输出,没有一个更新花费的时间大于running_updates_limit秒。 在线切换步骤如下: 首先,manager节点上停掉MHA监控: [root@Manager_Slave ~]# masterha_stop --conf=/etc/masterha/app1.cnf 其次,进行在线切换操作(模拟在线切换主库操作,原主库182.48.115.236变为slave,182.48.115.237提升为新的主库) [root@Manager_Slave ~]# masterha_master_switch --conf=/etc/masterha/app1.cnf --master_state=alive --new_master_host=182.48.115.237 --new_master_port=3306 --orig_master_is_new_slave --running_updates_limit=10000 ......... Thu Jun 1 00:28:02 2014 - [info] Executed CHANGE MASTER. Thu Jun 1 00:28:02 2014 - [info] Slave started. Thu Jun 1 00:28:02 2014 - [info] All new slave servers switched successfully. Thu Jun 1 00:28:02 2014 - [info] Thu Jun 1 00:28:02 2014 - [info] * Phase 5: New master cleanup phease.. Thu Jun 1 00:28:02 2014 - [info] Thu Jun 1 00:28:02 2014 - [info] 192.168.0.60: Resetting slave info succeeded. Thu Jun 1 00:28:02 2014 - [info] Switching master to 192.168.0.60(192.168.0.60:3306) completed successfully. 其中参数的意思: --orig_master_is_new_slave 切换时加上此参数是将原 master 变为 slave 节点,如果不加此参数,原来的 master 将不启动 --running_updates_limit=10000 故障切换时,候选master 如果有延迟的话, mha 切换不能成功,加上此参数表示延迟在此时间范围内都可切换(单位为s),但是切换的 时间长短是由recover 时relay 日志的大小决定 注意: 由于在线进行切换需要调用到master_ip_online_change这个脚本,但是由于该脚本不完整,需要进行相应的修改,脚本调整如下: [root@Manager_Slave ~]# vim /usr/local/bin/master_ip_online_change #!/usr/bin/env perl use strict; use warnings FATAL => 'all'; use Getopt::Long; use MHA::DBHelper; use MHA::NodeUtil; use Time::HiRes qw( sleep gettimeofday tv_interval ); use Data::Dumper; my $_tstart; my $_running_interval = 0.1; my ( $command, $orig_master_host, $orig_master_ip, $orig_master_port, $orig_master_user, $new_master_host, $new_master_ip, $new_master_port, $new_master_user, ); my $vip = '182.48.115.239/27'; # Virtual IP my $key = "1"; my $ssh_start_vip = "/sbin/ifconfig eth1:$key $vip"; my $ssh_stop_vip = "/sbin/ifconfig eth1:$key down"; my $ssh_user = "root"; my $new_master_password='123456'; my $orig_master_password='123456'; GetOptions( 'command=s' => \$command, #'ssh_user=s' => \$ssh_user, 'orig_master_host=s' => \$orig_master_host, 'orig_master_ip=s' => \$orig_master_ip, 'orig_master_port=i' => \$orig_master_port, 'orig_master_user=s' => \$orig_master_user, #'orig_master_password=s' => \$orig_master_password, 'new_master_host=s' => \$new_master_host, 'new_master_ip=s' => \$new_master_ip, 'new_master_port=i' => \$new_master_port, 'new_master_user=s' => \$new_master_user, #'new_master_password=s' => \$new_master_password, ); exit &main(); sub current_time_us { my ( $sec, $microsec ) = gettimeofday(); my $curdate = localtime($sec); return $curdate . " " . sprintf( "%06d", $microsec ); } sub sleep_until { my $elapsed = tv_interval($_tstart); if ( $_running_interval > $elapsed ) { sleep( $_running_interval - $elapsed ); } } sub get_threads_util { my $dbh = shift; my $my_connection_id = shift; my $running_time_threshold = shift; my $type = shift; $running_time_threshold = 0 unless ($running_time_threshold); $type = 0 unless ($type); my @threads; my $sth = $dbh->prepare("SHOW PROCESSLIST"); $sth->execute(); while ( my $ref = $sth->fetchrow_hashref() ) { my $id = $ref->{Id}; my $user = $ref->{User}; my $host = $ref->{Host}; my $command = $ref->{Command}; my $state = $ref->{State}; my $query_time = $ref->{Time}; my $info = $ref->{Info}; $info =~ s/^\s*(.*?)\s*$/$1/ if defined($info); next if ( $my_connection_id == $id ); next if ( defined($query_time) && $query_time < $running_time_threshold ); next if ( defined($command) && $command eq "Binlog Dump" ); next if ( defined($user) && $user eq "system user" ); next if ( defined($command) && $command eq "Sleep" && defined($query_time) && $query_time >= 1 ); if ( $type >= 1 ) { next if ( defined($command) && $command eq "Sleep" ); next if ( defined($command) && $command eq "Connect" ); } if ( $type >= 2 ) { next if ( defined($info) && $info =~ m/^select/i ); next if ( defined($info) && $info =~ m/^show/i ); } push @threads, $ref; } return @threads; } sub main { if ( $command eq "stop" ) { ## Gracefully killing connections on the current master # 1. Set read_only= 1 on the new master # 2. DROP USER so that no app user can establish new connections # 3. Set read_only= 1 on the current master # 4. Kill current queries # * Any database access failure will result in script die. my $exit_code = 1; eval { ## Setting read_only=1 on the new master (to avoid accident) my $new_master_handler = new MHA::DBHelper(); # args: hostname, port, user, password, raise_error(die_on_error)_or_not $new_master_handler->connect( $new_master_ip, $new_master_port, $new_master_user, $new_master_password, 1 ); print current_time_us() . " Set read_only on the new master.. "; $new_master_handler->enable_read_only(); if ( $new_master_handler->is_read_only() ) { print "ok.\n"; } else { die "Failed!\n"; } $new_master_handler->disconnect(); # Connecting to the orig master, die if any database error happens my $orig_master_handler = new MHA::DBHelper(); $orig_master_handler->connect( $orig_master_ip, $orig_master_port, $orig_master_user, $orig_master_password, 1 ); ## Drop application user so that nobody can connect. Disabling per-session binlog beforehand #$orig_master_handler->disable_log_bin_local(); #print current_time_us() . " Drpping app user on the orig master..\n"; #FIXME_xxx_drop_app_user($orig_master_handler); ## Waiting for N * 100 milliseconds so that current connections can exit my $time_until_read_only = 15; $_tstart = [gettimeofday]; my @threads = get_threads_util( $orig_master_handler->{dbh}, $orig_master_handler->{connection_id} ); while ( $time_until_read_only > 0 && $#threads >= 0 ) { if ( $time_until_read_only % 5 == 0 ) { printf "%s Waiting all running %d threads are disconnected.. (max %d milliseconds)\n", current_time_us(), $#threads + 1, $time_until_read_only * 100; if ( $#threads < 5 ) { print Data::Dumper->new( [$_] )->Indent(0)->Terse(1)->Dump . "\n" foreach (@threads); } } sleep_until(); $_tstart = [gettimeofday]; $time_until_read_only--; @threads = get_threads_util( $orig_master_handler->{dbh}, $orig_master_handler->{connection_id} ); } ## Setting read_only=1 on the current master so that nobody(except SUPER) can write print current_time_us() . " Set read_only=1 on the orig master.. "; $orig_master_handler->enable_read_only(); if ( $orig_master_handler->is_read_only() ) { print "ok.\n"; } else { die "Failed!\n"; } ## Waiting for M * 100 milliseconds so that current update queries can complete my $time_until_kill_threads = 5; @threads = get_threads_util( $orig_master_handler->{dbh}, $orig_master_handler->{connection_id} ); while ( $time_until_kill_threads > 0 && $#threads >= 0 ) { if ( $time_until_kill_threads % 5 == 0 ) { printf "%s Waiting all running %d queries are disconnected.. (max %d milliseconds)\n", current_time_us(), $#threads + 1, $time_until_kill_threads * 100; if ( $#threads < 5 ) { print Data::Dumper->new( [$_] )->Indent(0)->Terse(1)->Dump . "\n" foreach (@threads); } } sleep_until(); $_tstart = [gettimeofday]; $time_until_kill_threads--; @threads = get_threads_util( $orig_master_handler->{dbh}, $orig_master_handler->{connection_id} ); } print "Disabling the VIP on old master: $orig_master_host \n"; &stop_vip(); ## Terminating all threads print current_time_us() . " Killing all application threads..\n"; $orig_master_handler->kill_threads(@threads) if ( $#threads >= 0 ); print current_time_us() . " done.\n"; #$orig_master_handler->enable_log_bin_local(); $orig_master_handler->disconnect(); ## After finishing the script, MHA executes FLUSH TABLES WITH READ LOCK $exit_code = 0; }; if ($@) { warn "Got Error: $@\n"; exit $exit_code; } exit $exit_code; } elsif ( $command eq "start" ) { ## Activating master ip on the new master # 1. Create app user with write privileges # 2. Moving backup script if needed # 3. Register new master's ip to the catalog database # We don't return error even though activating updatable accounts/ip failed so that we don't interrupt slaves' recovery. # If exit code is 0 or 10, MHA does not abort my $exit_code = 10; eval { my $new_master_handler = new MHA::DBHelper(); # args: hostname, port, user, password, raise_error_or_not $new_master_handler->connect( $new_master_ip, $new_master_port, $new_master_user, $new_master_password, 1 ); ## Set read_only=0 on the new master #$new_master_handler->disable_log_bin_local(); print current_time_us() . " Set read_only=0 on the new master.\n"; $new_master_handler->disable_read_only(); ## Creating an app user on the new master #print current_time_us() . " Creating app user on the new master..\n"; #FIXME_xxx_create_app_user($new_master_handler); #$new_master_handler->enable_log_bin_local(); $new_master_handler->disconnect(); ## Update master ip on the catalog database, etc print "Enabling the VIP - $vip on the new master - $new_master_host \n"; &start_vip(); $exit_code = 0; }; if ($@) { warn "Got Error: $@\n"; exit $exit_code; } exit $exit_code; } elsif ( $command eq "status" ) { # do nothing exit 0; } else { &usage(); exit 1; } } # A simple system call that enable the VIP on the new master sub start_vip() { `ssh $ssh_user\@$new_master_host \" $ssh_start_vip \"`; } # A simple system call that disable the VIP on the old_master sub stop_vip() { `ssh $ssh_user\@$orig_master_host \" $ssh_stop_vip \"`; } sub usage { print "Usage: master_ip_online_change --command=start|stop|status --orig_master_host=host --orig_master_ip=ip --orig_master_port=port --new_master_host=host --new_master_ip=ip --new_master_port=port\n"; die; }
由于在线进行切换需要调用到master_ip_online_change这个脚本,但是由于该脚本不完整,需要自己进行相应的修改。在测试中发现脚本还是有问题,脚本中new_master_password这个变量获取不到,导致在线切换失败,所以进行了相关的硬编码,直接把mysql的root用户密码赋值给变量new_master_password。另外这个脚本还可以管理vip。
10. 修复宕机后的master节点
通常情况下自动切换以后,原master可能已经废弃掉,待原master主机修复后,如果数据完整的情况下,可能想把原来master重新作为新主库的slave,这时我们可以借助当时自动切换时刻的MHA日志来完成对原master的修复。下面是提取相关日志的命令:
[root@Manager_Slave ~]# grep -i "All other slaves should start" manager.log Mon Apr 21 22:28:33 2016 - [info] All other slaves should start replication from here. Statement should be: CHANGE MASTER TO MASTER_HOST='182.48.115.237', MASTER_PORT=3306, MASTER_LOG_FILE='mysql-bin.000022', MASTER_LOG_POS=506716, MASTER_USER='repl', MASTER_PASSWORD='xxx';
获取上述信息以后,就可以直接在修复后的master上执行change master to相关操作,重新作为从库了。
最后补充一下邮件发送脚本send_report ,这个脚本经过调整后可以使用,如下:
#!/usr/bin/perl # Copyright (C) 2011 DeNA Co.,Ltd. # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation; either version 2 of the License, or # (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., # 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA ## Note: This is a sample script and is not complete. Modify the script based on your environment. use strict; use warnings FATAL => 'all'; use Mail::Sender; use Getopt::Long; #new_master_host and new_slave_hosts are set only when recovering master succeeded my ( $dead_master_host, $new_master_host, $new_slave_hosts, $subject, $body ); my $smtp='smtp.163.com'; my $mail_from='xxxx'; my $mail_user='xxxxx'; my $mail_pass='xxxxx'; my $mail_to=['xxxx','xxxx']; GetOptions( 'orig_master_host=s' => \$dead_master_host, 'new_master_host=s' => \$new_master_host, 'new_slave_hosts=s' => \$new_slave_hosts, 'subject=s' => \$subject, 'body=s' => \$body, ); mailToContacts($smtp,$mail_from,$mail_user,$mail_pass,$mail_to,$subject,$body); sub mailToContacts { my ( $smtp, $mail_from, $user, $passwd, $mail_to, $subject, $msg ) = @_; open my $DEBUG, "> /tmp/monitormail.log" or die "Can't open the debug file:$!\n"; my $sender = new Mail::Sender { ctype => 'text/plain; charset=utf-8', encoding => 'utf-8', smtp => $smtp, from => $mail_from, auth => 'LOGIN', TLS_allowed => '0', authid => $user, authpwd => $passwd, to => $mail_to, subject => $subject, debug => $DEBUG }; $sender->MailMsg( { msg => $msg, debug => $DEBUG } ) or print $Mail::Sender::Error; return 1; } # Do whatever you want here exit 0;
告警邮件如下:
目前高可用方案可以一定程度上实现数据库的高可用,出于对数据库的高可用和数据一致性的要求,推荐使用MHA架构。