MHA – GTID模式

MasterHigh Availability Manager and Toolsfor MySQL.是采用PERL语音编写的一个脚本管理工具，该工具适用于MySQL Replication环境，目的是在与维持master主库的高可用性。

MHA是自动的master故障转移和slave提升的软件包，基于标准的MySQL复制(异步、半同步)

MHA Manager管理节点可以单独部署在一个独立服务器上管理多个master-slave集群，也可以部署在一台slave上。MHA Manager探测集群中的node节点

MHA有两个组成部分。

MHA Manager (管理节点)

MHA node (数据节点)

下载地址：

https://github.com/yoshinorim/mha4mysql-manager/releases

https://github.com/yoshinorim/mha4mysql-node/releases

MHA Manager管理节点可以单独部署在一台服务器上管理多个master-slave集群，也可以部署在一台slave上，MHA Manager探测集群中的node节点，当发现master出现故障时，它可以自动将具有最新数据的slave提升为新master,然后将所有其他slave导向新的master上。

整个故障转移过程对应用程序是透明的，MHA node数据节点可以运行在每台Mysql服务器上，

MHA的目的是维持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，然后从中选择出一个充当新的master,并将其他slave指向它。当master出现故障时，可以通过对比slave之间I/O thread读取主库binlog的position号，选取最接近的slave作为备选主库，其他的从库可以通过与备选主库对比生成差异的中继日志，在备选主库上应用从原来master保存的binlog,同时将备选主库提升为master,最后在其他slave上应用相应的差异中继日志并从新master开始复制。

环境配置

\

主服务器1

从服务器1

从服务器2

OS

Centos el7.x86_64

Centos el7.x86_64

Centos el7.x86_64

IP

192.168.31.79

192.168.31.188

192.168.31.90

HOSTNAME

mysql

Mysql2

Mysql3

Mysql-version

5.7.23

5.7.23

5.7.23

数据库

TEST

备注1

vip 192.168.31.100

备注2

管理节点

备注3

候选节点

备注4

单网卡 名：enp0s3

单网卡 名：enp0s3

单网卡 名：enp0s3

三个节点安装mysql

同时，在主节点创建数据库TEST及TEST数据库下建立若干表，用于测试目的。

三个节点均运行

建立复制账号

mysql> create user 'rep1'@'192.168.31.%' identified by 'Oracle123';

mysql> grant replication slave on *.* to 'rep1'@'192.168.31.%';

mysql> flush privileges;

建立管理账号

mysql> create user 'dba'@'192.168.31.%' identified by 'Oracle123';

mysql> grant all privileges on *.* to 'dba'@'192.168.31.%';

mysql> flush privileges;

主服务器1

[root@mysql ~]# vi /etc/my.cnf

添加内容如下

gtid_mode=on

enforce_gtid_consistency=on

log_bin=on

binlog_format=row

server-id=79

log-bin = master-log

relay-log = relay-log

skip_name_resolve

从服务器1

[root@mysql2 ~]# vi /etc/my.cnf

gtid_mode=on

enforce_gtid_consistency=on

log_slave_updates=1

server-id=188

relay-log = relay-log

log-bin = master-log

skip_name_resolve

relay_log_purge = 0

##read_only = ON该从库会被选为候选主库，故取消read_only

从服务器2

[root@mysql3 ~]# vi /etc/my.cnf

gtid_mode=on

enforce_gtid_consistency=on

log_slave_updates=1

server-id=90

relay-log = relay-log

log-bin = master-log

skip_name_resolve

relay_log_purge = 0

read_only = ON

主服务器1

备份TEST数据库

[root@mysql ~]# mysqldump --single-transaction -uroot -pOracle123 TEST > TEST_20200316.sql

从服务器1

[root@mysql2 ~]# scp root@192.168.31.79:/root/TEST_20200316.sql .

mysql> create database TEST;

[root@mysql2 ~]# mysql -uroot -pOracle123 TEST < TEST_20200316.sql

从服务器2

[root@mysql3 ~]# scp root@192.168.31.79:/root/TEST_20200316.sql .

mysql> create database TEST;

[root@mysql3 ~]# mysql -uroot -pOracle123 TEST < TEST_20200316.sql

三个节点都重启

[root@mysql ~]# service mysql stop

[root@mysql ~]# service mysql start

从服务器两个节点都执行

[root@mysql2 bin]# mysql -uroot -p

mysql> change master to master_host='192.168.31.79',master_user='rep1',master_password='Oracle123',master_port=3306,master_auto_position=1;

mysql> start slave;

mysql> show slave status\G

注：如果以上执行change master有报错，Last_IO_Error: Got fatal error 1236 from master when reading data from binary log: 'The slave is connecting using CHANGE MASTER TO MASTER_AUTO_POSITION = 1, but the master has purged binary logs containing GTIDs that the slave requires.'

解决方式如下：

mysql> change master to master_auto_position=0;

mysql> start slave;

注：如果执行以上出现报错，

Last_Error: Error 'Table 'test4' already exists' on query. Default database: 'TEST'. Query: 'create table test4 (id integer,name varchar(200))'

解决方式如下:

主库：mysql> SHOW MASTER STATUS\g --获取File Position

从库执行

mysql> change master to master_host='192.168.31.79',master_user='rep1', master_password='Oracle123',master_port=3306,master_log_file='master-log.000001',master_log_pos=154;

主服务器1

[root@mysql ~]# mysql -uroot -p

mysql> show master status;

主服务器1

mysql> use TEST;

mysql> insert into test2 values (1,'test1');

mysql> commit;

从服务器1

mysql> use TEST;

mysql> select * from test2;

从服务器2

mysql> use TEST;

mysql> select * from test2;

三个节点均执行(包含管理节点)

[root@mysql ~]# cd .ssh/

[root@mysql .ssh]# ssh-keygen -t dsa -P '' -f id_dsa

[root@mysql .ssh]# cat id_dsa.pub >> authorized_keys

主服务器接收秘钥

主服务器操作

[root@mysql .ssh]# scp 192.168.31.188:/root/.ssh/id_dsa.pub ./id_dsa.pub_188

[root@mysql .ssh]# scp 192.168.31.90:/root/.ssh/id_dsa.pub ./id_dsa.pub_90

[root@mysql .ssh]# cat id_dsa.pub_188 >> authorized_keys

[root@mysql .ssh]# cat id_dsa.pub_90 >> authorized_keys

[root@mysql .ssh]# scp authorized_keys 192.168.31.188:/root/.ssh/

[root@mysql .ssh]# scp authorized_keys 192.168.31.90:/root/.ssh/

三个节点配置host

三个节点均执行

[root@mysql ~]# vi /etc/hosts

添加内如如下:

192.168.31.79 mysql

192.168.31.188 mysql2

192.168.31.90 mysql3

验证

三个节点均互相测试,不需要输入密码即成功

[root@mysql .ssh]# ssh mysql2

[root@mysql .ssh]# ssh mysql3

NODE节点安装

[root@mysql soft]# yum install -y perl-CPAN

[root@mysql soft]# yum install -y mha4mysql-node-0.58-0.el7.centos.noarch.rpm

管理节点安装(即从服务器1)

如果缺少包，可以下载之后在安装

https://centos.pkgs.org/

[root@mysql2 ~]# yum install perl-Config-Tiny perl-Email-Date-Format perl-File-Remove perl-Log-Dispatch perl-Mail-Sender perl-Mail-Sendmail perl-MIME-Lite perl-MIME-Types perl-Module-Install perl-Module-ScanDeps perl-Parallel-ForkManager perl-YAML -y

[root@mysql2 soft]# yum install mha4mysql-manager-0.58-0.el7.centos.noarch.rpm

每一个节点

[root@mysql ~]# cd /usr/local/

[root@mysql local]# mkdir mha

管理节点(从服务器1)

[root@mysql2 soft]# mkdir /usr/local/mha

[root@mysql2 mha]# vi mha.conf

添加内容如下：

[server default]

user = dba

password = Oracle123

ssh_user = root

repl_user = rep1

repl_password = Oracle123

ping_interval = 1

ping_type = SELECT

manager_workdir=/usr/local/mha

manager_log=/usr/local/mha/manager.log

remote_workdir=/usr/local/mha

master_ip_failover_script="/usr/local/mha/master_ip_failover"

master_ip_online_change_script="/usr/local/mha/master_ip_online_change"

shutdown_script=""

report_script=""

#check_repl_delay=0

[server1]

hostname=mysql

port=3306

master_binlog_dir="/u01/data"

candidate_master=1##优先选择该节点为候选节点

[server2]

hostname=mysql2

port=3306

master_binlog_dir="/u01/data"

candidate_master=1

[server3]

hostname=mysql3

port=3306

master_binlog_dir="/u01/data"

no_master =1##不选择该节点为候选节点

[root@mysql2 mha]# vi master_ip_failover

添加内容如下

#!/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 = '192.168.31.100/32'; # Virtual IP，该ip具体参考步骤12添加IP，显示出的结果

my $key = "0";

my $int = "enp0s3";

##my $ssh_start_vip = "/sbin/ifconfig $int:$key $vip"; --本环境为centos7，故注释该指令

##my $ssh_stop_vip = "/sbin/ifconfig $int:$key down"; --本环境为centos7，该指令无效

my $ssh_start_vip = "/sbin/ip addr add $vip dev $int";

my $ssh_stop_vip = "/sbin/ip addr del $vip dev $int";

my $arp_effect = "/sbin/arping -Uq-s192.168.31.100 -I $int 192.168.31.255-c 3"; # Virtual IP and gatway

#my $test = "echo successfull >/tmp/test.txt";

$ssh_user = "root";

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" ) {

# $orig_master_host, $orig_master_ip, $orig_master_port are passed.

# If you manage master ip address at global catalog database,

# invalidate orig_master_ip here.

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" ) {

# all arguments are passed.

# If you manage master ip address at global catalog database,

# activate new_master_ip here.

# You can also grant write access (create user, set read_only=0, etc) here.

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 \"`;

&status();

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 \"`;

`ssh $ssh_user\@$new_master_host \" $arp_effect \"`;

# `ssh $ssh_user\@$new_master_host \" $test \"`;

}

# 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 status() {

print `ssh $ssh_user\@$orig_master_host \" ip add show $int \"`;

}

sub usage {

print

"Usage: master_ip_failover --command=start|stop|stopssh|status --orig_master_host=host --orig_maste

r_ip=ip --orig_master_port=port --new_master_host=host --new_master_ip=ip --new_master_port=port\n";

}

[root@mysql2 mha]# vi master_ip_online_change

添加内容如下：

#!/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 $vip = "192.168.31.100/32";

my $if = "enp0s3";

my (

$command,$orig_master_is_new_slave,$orig_master_host,

$orig_master_ip,$orig_master_port,$orig_master_user,

$orig_master_password, $orig_master_ssh_user,$new_master_host,

$new_master_ip,$new_master_port,$new_master_user,

$new_master_password, $new_master_ssh_user,

);

GetOptions(

'command=s'=> \$command,

'orig_master_is_new_slave'=> \$orig_master_is_new_slave,

'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,

'orig_master_ssh_user=s'=> \$orig_master_ssh_user,

'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,

'new_master_ssh_user=s' => \$new_master_ssh_user,

);

exit &main();

sub drop_vip {

my $output = `ssh -o ConnectTimeout=15 -o ConnectionAttempts=3 $orig_master_host /sbin/ip addr del $vip/32 dev $if`;

}

sub add_vip {

my $output = `ssh -o ConnectTimeout=15 -o ConnectionAttempts=3 $new_master_host /sbin/ip addr add $vip/32 dev $if`;

}

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";

print current_time_us() . " drop vip $vip..\n";

#drop_app_user($orig_master_handler);

&drop_vip();

## 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} );

}

## 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";

print current_time_us() . "Add vip $vip on $if..\n";

# create_app_user($new_master_handler);

&add_vip();

$new_master_handler->enable_log_bin_local();

$new_master_handler->disconnect();

## Update master ip on the catalog database, etc

$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;

}

}

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;

}

管理节点

[root@mysql2 mha]# which masterha_check_ssh

[root@mysql2 mha]# chmod +x master_ip_online_change

[root@mysql2 mha]# chmod +x master_ip_failover

[root@mysql2 mha]# masterha_check_ssh --conf=/usr/local/mha/mha.conf

[root@mysql2 mha]# masterha_check_repl --conf=/usr/local/mha/mha.conf

主服务器1

[root@mysql ~]# ip addr add 192.168.31.100 dev enp0s3

[root@mysql ~]# ip addr show

注意是 100/32

管理节点(即从服务器1)

[root@mysql2 mha]# nohup masterha_manager --conf=/usr/local/mha/mha.conf > /tmp/mha_manager.log < /dev/null 2>&1 &

管理节点(即从服务器1)

[root@mysql2 mha]# masterha_check_status --conf=/usr/local/mha/mha.conf

主服务器1

[root@mysql ~]# service mysql stop

注：

每次出发MHA切换,需要在管理节点删除mha.failover.complete文件，不然下次切换会报错。

从服务器1(候选节点)

检查

[root@mysql2 mha]# tail -100f manager.log

[root@mysql2 mha]# ip addr show

mysql> show slave status\G

mysql> show master status\G

从服务器2

将原主库恢复，成为现集群环境的备库

现主库操作

mysql> show master status;

原主库操作

[root@mysql data]# service mysql start

[root@mysql data]# mysql -uroot -p

mysql> change master to master_host='192.168.31.188',master_user='rep1', master_password='Oracle123',master_port=3306,master_log_file='master-log.000001',master_log_pos=409;

mysql> start slave;

mysql> show slave status\G