MHA部署筆記(centos6+mysql5.6)

環(huán)境及軟件版本：

    CentOS6.5x86_64

    MySQL5.6.34編譯安裝版

    MHA版本：mha4mysql-manager-0.56-0.el6.noarch.rpm mha4mysql-node-0.56-0.el6.noarch.rpm

節(jié)點(diǎn)角色：

    node93：10.1.20.93   默認(rèn)主庫

    node94：10.1.20.94   從庫1，原先主庫宕機(jī)后可提升為主庫【mha管理節(jié)點(diǎn)也部署在這臺機(jī)器上】

    node95：10.1.20.95   從庫2，不允許提升為主庫

   準(zhǔn)備的VIP是 10.1.20.100/24

step1、配置主從關(guān)系

node93的/etc/my.cnf配置文件部分關(guān)鍵地方：

[mysqld]

port            = 3306

socket          = /tmp/mysql.sock

datadir         = /bdata/data/nowdb2

innodb_file_per_table=ON

character-set-server = utf8

default_storage_engine = InnoDB

skip-innodb_adaptive_hash_index

master_info_repository = TABLE

relay_log_info_repository = TABLE

relay_log_recovery = 1 #crash safe

log-bin=mysql-bin

binlog_format=row

sync_binlog =1  #確保事務(wù)提交的時(shí)候BINLOG落盤

log-slave-updates

log_bin_trust_function_creators =1

binlog_rows_query_log_events=ON  #記錄執(zhí)行的語句到BINLOG query event

server-id=1020093

relay_log_purge=0

read_only=1

node94的/etc/my.cnf配置文件部分關(guān)鍵地方：

[mysqld]

port            = 3306

socket          = /tmp/mysql.sock

datadir         = /bdata/data/nowdb2

innodb_file_per_table=ON

character-set-server = utf8

default_storage_engine = InnoDB

skip-innodb_adaptive_hash_index

master_info_repository = TABLE

relay_log_info_repository = TABLE

relay_log_recovery = 1 #crash safe

log-bin=mysql-bin

binlog_format=row

sync_binlog =1  #確保事務(wù)提交的時(shí)候BINLOG落盤

log-slave-updates

log_bin_trust_function_creators =1

binlog_rows_query_log_events=ON  #記錄執(zhí)行的語句到BINLOG query event

server-id=1020094

relay_log_purge=0

read_only=1

node95的/etc/my.cnf配置文件部分關(guān)鍵地方：

[mysqld]

port            = 3306

socket          = /tmp/mysql.sock

datadir         = /bdata/data/nowdb2

innodb_file_per_table=ON

character-set-server = utf8

default_storage_engine = InnoDB

skip-innodb_adaptive_hash_index

master_info_repository = TABLE

relay_log_info_repository = TABLE

relay_log_recovery = 1 #crash safe

log-bin=mysql-bin

binlog_format=row

sync_binlog =1  #確保事務(wù)提交的時(shí)候BINLOG落盤

log-slave-updates

log_bin_trust_function_creators = 1

binlog_rows_query_log_events=ON  #記錄執(zhí)行的語句到BINLOG query event

server-id=1020095

relay_log_purge=0

read_only=1

在node93上創(chuàng)建復(fù)制權(quán)限的賬號，GRANT REPLICATION SLAVE ,REPLICATION CLIENT ON *.* TO 'rpl'@'10.1.%.%' IDENTIFIED BY 'rpl';

然后配置1主2從，（具體步驟略過）。

注意：我們要確保能成為主庫的節(jié)點(diǎn)(node93、node94)都存在主從同步賬號，如果node94上不存在rpl賬號，就到node94節(jié)點(diǎn)去手工添加即可。

主從關(guān)系建立好后，我們在master上創(chuàng)建個(gè)mha管理賬號，后期會用到：

grant all on *.* to 'mhauser'@'10.1.%.%' identified by 'Abcd@1234';   

(管理賬號要在node93、node94、node95所有節(jié)點(diǎn)都存在)

step2、安裝MHA

因?yàn)?/span>MHA依賴于SSH，因此需要在3臺主機(jī)之間建立SSH免秘鑰登陸。步驟略過。

3個(gè)節(jié)點(diǎn)都安裝perl包：

yum install perl perl-DBD-MySQL perl-CPAN perl-devel perl-Time-HiRes

node93-node95上都安裝node包：

rpm -ivh mha4mysql-node-0.56-0.el6.noarch.rpm

node94上安裝Manager包(當(dāng)然，我們在3個(gè)節(jié)點(diǎn)都安裝上Manager包也沒問題)：

rpm -ivh mha4mysql-manager-0.56-0.el6.noarch.rpm

在node94初始化MHA

mkdir /etc/masterha/

vim /etc/masterha/app1.cnf 內(nèi)容如下：

[server default]

user=mhauser

password=Abcd@1234

manager_workdir=/data/masterha/app1

manager_log=/data/masterha/app1/manager.log

remote_workdir=/data/masterha/app1

ssh_user=root

repl_user=rpl

repl_password=rpl

ping_interval=1

master_binlog_dir=/bdata/data/nowdb2/    # 這個(gè)路徑和你的mysql的binlog存放路徑要一致

master_ip_failover_script=/etc/masterha/master_ip_failover

report_script=/etc/masterha/send_report

#通過第三方機(jī)器確認(rèn)目標(biāo)主庫是否存活,不是必須的,就算沒有也是能用

#secondary_check_script=masterha_secondary_check -s remote_host1 -s remote_host2

#故障發(fā)生后關(guān)閉主機(jī)的腳本,不是必須的,但是你要設(shè)置為空

# shutdown_script=""

#手動(dòng)在線切換VIP腳本,不是必須的,就算沒有也是能用,

#如果你有keepalived這種來做切換VIP就可以直接不用了

master_ip_online_change_script==/etc/masterha/master_ip_online_change

[server1]

hostname=10.1.20.93

candidate_master=1

[server2]

hostname=10.1.20.94

candidate_master=1

[server3]

hostname=10.1.20.95

no_master=1   # 禁止

在node94上添加腳本/etc/masterha/master_ip_failover  (里面填寫上相關(guān)的VIP的信息)

#!/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,

   $orig_master_ssh_port, $new_master_ssh_port

);

my $vip ='10.1.20.100';  # Virtual IP

my $devic='eth0';

my $key = "0";

my$net_mask='255.255.255.0';

my $ssh_start_vip ="/sbin/ifconfig $devic:$key $vip netmask $net_mask";

my $ssh_stop_vip ="/sbin/ifconfig $devic:$key down";

my$mysql_conf="/etc/my.cnf";

my $open_readonly="/bin/sed-i 's/.*read_only.*/read_only=1/g' $mysql_conf ";

my $close_readonly="/bin/sed-i 's/.*read_only.*/read_only=0/g' $mysql_conf ";

my $open_relaylog_purge="/bin/sed-i 's/.*relay_log_purge.*/relay_log_purge=0/g' $mysql_conf ";

my$close_relaylog_purge="/bin/sed -i's/.*relay_log_purge.*/relay_log_purge=0/g' $mysql_conf ";

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_ssh_port=i' => \$orig_master_ssh_port,

   'new_master_host=s'  => \$new_master_host,

   'new_master_ip=s'    =>\$new_master_ip,

   'new_master_port=i'  =>\$new_master_port,

   'new_master_ssh_port=i' => \$new_master_ssh_port,

);

exit &main();

sub main {

   print "\n\nIN SCRIPTTEST====$ssh_stop_vip==$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 atglobal catalog database,

        # invalidate orig_master_ip here.

        my $exit_code = 1;

        eval {

            print "Disabling the VIP onold 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 atglobal 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 - $vipon 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 thescript.. OK \n";

       # `ssh $ssh_user\@cluster1 \"$ssh_start_vip \"`;

        exit 0;

   }

   else {

        &usage();

        exit 1;

   }

}

# A simple system call that enablethe VIP on the new master

sub start_vip() {

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

   print "Disable read_only and relay_log_purge in my.cnf -  on the new master - $new_master_host\n";

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

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

}

# A simple system call thatdisable the VIP on the old_master

sub stop_vip() {

   `ssh $ssh_user\@$orig_master_host \" $ssh_stop_vip \"`;

  print "Enable read_only and relay_log_purge in my.cnf -  on the orig master - $orig_master_host\n";

   `ssh $ssh_user\@$orig_master_host \" $open_readonly \"`;

   `ssh $ssh_user\@$orig_master_host \" $open_relaylog_purge \"`;

}

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--orig_master_ssh_port=ssh_port --new_master_ssh_port = ssh_port\n";

}

在node94上添加腳本/etc/masterha/send_report(里面填寫上相關(guān)的smtp賬號的信息)：

#!/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 scriptand 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_hostsare set only when recovering master succeeded

my ( $dead_master_host,$new_master_host, $new_slave_hosts, $subject, $body ,$conf);

my$smtp='smtp.exmail.qq.com';

my$mail_from='zabbix@xxxx.com';

my$mail_user='zabbix@xxxx.com';

my $mail_pass='xxxxxxx';

my$mail_to=['lixxxx@xxxx.com'];

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,

 'conf=s'             => \$conf,

);

mailToContacts($smtp,$mail_from,$mail_user,$mail_pass,$mail_to,$subject,$body);

check_if_sendmail_ok('/tmp/monitormail.log');

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;

}

sub check_if_sendmail_ok{

    #>>250 2.0.0 Ok: queued as 3532C6DA009D

    #<<QUIT

    #>>221 2.0.0 Bye

    my$logf = shift;

    openRLOG, $logf or die "cannot open file $logf.\n";

    my@log = <RLOG>;

    closeRLOG;

    my$val = 0;

    if($log[$#log]=~ m/\>\>\s221\s.*\sBye/){

       print"Meet Bye.\t";

       $val++;

    }

    if($log[$#log-1]=~ m/\<\<\sQUIT/){

       print"Meet QUIT.\t";

       $val++;

    }

    if($log[$#log-2]=~ m/\>\>\s250\s.*\sOk: queued/){

       print"Meet queued.\t";

       $val++;

    }

    print"\n";

    if($val== 3){

       print"send mail success.\n";

    }

    else{

       print"send mail failed.check DNS/SMTP config\n";

    }

    return$val;

}

# Do whatever you want here

exit 0;

在node94上添加腳本/etc/masterha/master_ip_online_change(里面填寫上相關(guān)的VIP的信息)：

#!/usr/bin/env 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 scriptand is not complete. Modify the script based on your environment.

use strict;

use warnings FATAL => 'all';

use Getopt::Long;

use MHA::DBHelper;

use MHA::NodeUtil;

use Time::HiRes qw( sleepgettimeofday tv_interval );

use Data::Dumper;

my $_tstart;

my $_running_interval = 0.1;

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

);

my $vip ='10.1.20.100/24';

my $key = '0';

my $ssh_start_vip ="/sbin/ifconfig eth0:$key $vip";

my $ssh_stop_vip ="/sbin/ifconfig eth0:$key down";

my $orig_master_ssh_port = 22;

my $new_master_ssh_port = 22;

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,

 'orig_master_ssh_port=i'    =>\$orig_master_ssh_port,

 'new_master_ssh_port=i'    =>\$new_master_ssh_port,

);

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 = newMHA::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() . " Setread_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 ifany database error happens

      my $orig_master_handler = newMHA::DBHelper();

     $orig_master_handler->connect($orig_master_ip, $orig_master_port,

        $orig_master_user,$orig_master_password, 1 );

      ## Drop application user so that nobodycan connect. Disabling per-session binlog beforehand

      $orig_master_handler->disable_log_bin_local();

      print current_time_us() . " Drppingapp user on the orig master..\n";

     #FIXME_xxx_drop_app_user($orig_master_handler);

      ## Waiting for N * 100 milliseconds sothat 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 %dthreads are disconnected.. (max %d milliseconds)\n",

            current_time_us(), $#threads + 1,$time_until_read_only * 100;

          if ( $#threads < 5 ) {

           printData::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 currentmaster so that nobody(except SUPER) can write

      print current_time_us() . " Setread_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 sothat 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 %dqueries 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() . " Killingall 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, MHAexecutes FLUSH TABLES WITH READ LOCK

      eval {

      `ssh -p$orig_master_ssh_port$orig_master_ssh_user\@$orig_master_host \" $ssh_stop_vip \"`;

        };

        if ($@) {

            warn $@;

        }

      $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 eventhough activating updatable accounts/ip failed so that we don't interruptslaves' recovery.

# If exit code is 0 or 10, MHAdoes not abort

   my $exit_code = 10;

   eval {

      my $new_master_handler = newMHA::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() . " Setread_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() . " Creatingapp 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 catalogdatabase, etc

      `ssh -p$new_master_ssh_port$new_master_ssh_user\@$new_master_host \" $ssh_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;

 }

}

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;

}

在node94上檢查MHA 的SSH是否配置無誤：

    masterha_check_ssh--conf=/etc/masterha/app1.cnf

    如能出現(xiàn)"All SSH connection tests passedsuccessfully." 說明配置沒問題

在node94上檢查MHA 的主從復(fù)制是否配置無誤：

    masterha_check_repl--conf=/etc/masterha/app1.cnf

    如提示"MySQL Replication Health is OK"說明配置沒問題

在node94上前臺啟動(dòng)MHA：

    masterha_manager--conf=/etc/masterha/app1.cnf --ignore_last_failover  在前臺啟動(dòng)監(jiān)控

模擬node93master宕機(jī)、觀察master的自動(dòng)切換：

將node93的mysql服務(wù)停掉，可以發(fā)現(xiàn)此時(shí)node94上開啟的masterha_manager進(jìn)程自動(dòng)退出了，到其他節(jié)點(diǎn)去查看，可以發(fā)現(xiàn)主從切換了。

然后啟動(dòng)node93的mysql，再次上線就不會自動(dòng)變成master

【!!注意：直接將node93上線的話，集群中就出現(xiàn)了2個(gè)主節(jié)點(diǎn)，腦裂，masterha_manger也無法啟動(dòng)】，我們需要先手工將其改為從節(jié)點(diǎn)，操作如下：

在node93上，執(zhí)行：

> change master to

    master_host='10.1.20.94',

    master_user='rpl',

    master_password='rpl',

    master_log_file='mysql-bin.000003',

    master_log_pos=1881;     # 這里的位置，需要看下node94的/data/masterha/app1目錄下的manager.log里面的內(nèi)容找到具體的binlog位置。

> start slave;

> show slave status\G

將node93 重新加入集群后，我們在node94 manager節(jié)點(diǎn)上再次執(zhí)行 masterha_manager--conf=/etc/masterha/app1.cnf --ignore_last_failover  發(fā)現(xiàn)啟動(dòng)不退出了。(驗(yàn)證過程中不要關(guān)閉這個(gè)窗口)

檢查下當(dāng)前主從的配置：

    node94另外開一個(gè)xshell窗口，可以執(zhí)行 masterha_check_repl--conf=/etc/masterha/app1.cnf

    可以看出主節(jié)點(diǎn)、從節(jié)點(diǎn)的發(fā)生了變化:

查看是否masterha啟動(dòng)：

    另外開一個(gè)xshell窗口，可以執(zhí)行 masterha_check_status--conf=/etc/masterha/app1.cnf

如果需要停止masterha的話，不要用stop或者kill，要用下面的命令：

    masterha_stop--conf=/etc/masterha/app1.cnf

手動(dòng)切換主從的方法：

    masterha_master_switch-h   查看幫助信息

    masterha_master_switch--conf=/etc/masterha/app1.cnf --master_state=alive --new_master_host=10.1.20.93--new_master_port=3306 --orig_master_is_new_slave --running_updates_limit=10000

手工切換的時(shí)候需要注意2點(diǎn):

1、執(zhí)行手工切換的時(shí)候，需要先關(guān)掉老的master和即將提升為master的主機(jī)的 event scheduler，否則無法切換(set global event_scheduler = OFF;)

2、執(zhí)行手工切換的時(shí)候，需要先關(guān)閉MHA的監(jiān)控 masterha_stop--conf=/etc/masterha/app1.cnf )

3、執(zhí)行手工切換腳本的時(shí)候，它會自動(dòng)在原先的master上執(zhí)行FLUSH TABLES WITH READ LOCK; 等切換完成后，再UNLOCK TABLES釋放掉這個(gè)原有master的鎖。

發(fā)送郵件的腳本，需要先安裝插件：

    yuminstall perl-Mail-Sender

    發(fā)送失敗的話，可以查看/tmp/monitormail.log 找找失敗的原因。

MHA異常的話：

    可以查看日志路徑：/data/masterha/app1/

masterha_manager 還有幾個(gè)比較有用的啟動(dòng)參數(shù)：

    --remove_dead_master_conf    該參數(shù)代表當(dāng)發(fā)生主從切換后，老的主庫的ip將會從配置文件中移除。

    --manger_log    日志存放位置,想規(guī)范化管理日志可以加上

    --ignore_last_failover    該參數(shù)代表忽略上次MHA觸發(fā)切換產(chǎn)生的文件，默認(rèn)情況下，MHA發(fā)生切換后會在日志目錄，也就是上面我設(shè)置的/data產(chǎn)生app1.failover.complete文件，下次再次切換的時(shí)候如果發(fā)現(xiàn)該目錄下存在該文件將不允許觸發(fā)切換，除非在第一次切換后收到刪除該文件，在缺省情況下，如果MHA檢測到連續(xù)發(fā)生宕機(jī)，且兩次宕機(jī)間隔不足8小時(shí)的話，則不會進(jìn)行Failover，之所以這樣限制是為了避免ping-pong效應(yīng)?！救绻覀冃枰獜?qiáng)制切換的話，需要先移除這個(gè)文件app1.failover.complete】