Chapter X high availability to separate read and write

Master-slave replication architecture evolved introduction

basic structure

（1）一主一从
​（2）一主多从
​（3）多级主从
​（4）双主
​（5）循环复制

Advanced Application Architecture Evolution

High-performance architecture

(1) 读写分离架构(读性能较高)
代码级别
MySQL proxy (Atlas,mysql router,proxySQL(percona),maxscale)、amoeba(taobao) ,xx-dbproxy等。

(2) 分布式架构(读写性能都提高):
分库分表——cobar--->TDDL(头都大了),DRDS
Mycat--->DBLE自主研发等。
NewSQL-->TiDB

High availability architecture

（1）单活:MMM架构——mysql-mmm（google）
（2）单活:MHA架构——mysql-master-ha（日本DeNa）,T-MHA
（3）多活:MGR ——5.7 新特性 MySQL Group replication(5.7.17) --->Innodb Cluster  
（4）多活:MariaDB Galera Cluster架构,(PXC)Percona XtraDB Cluster、MySQL Cluster(Oracle rac)架构

Availability MHA *****

Architecture Works

#主库宕机处理过程

1. 监控节点 (通过配置文件获取所有节点信息)
   系统,网络,SSH连接性
   主从状态,重点是主库
2. 选主
(1) 如果判断从库(position或者GTID),数据有差异,最接近于Master的slave,成为备选主
(2) 如果判断从库(position或者GTID),数据一致,按照配置文件顺序,选主.
(3) 如果设定有权重(candidate_master=1),按照权重强制指定备选主.
    1. 默认情况下如果一个slave落后master 100M的relay logs的话，即使有权重,也会失效.
    2. 如果check_repl_delay=0的化,即使落后很多日志,也强制选择其为备选主
3. 数据补偿
(1) 当SSH能连接,从库对比主库GTID 或者position号,立即将二进制日志保存至各个从节点并且应用(save_binary_logs )
(2) 当SSH不能连接, 对比从库之间的relaylog的差异(apply_diff_relay_logs) 
4. Failover
将备选主进行身份切换,对外提供服务
其余从库和新主库确认新的主从关系
5. 应用透明(VIP)
6. 故障切换通知(send_reprt)
7. 二次数据补偿(binlog_server)
8. 自愈自治(待开发...)


# Failover

1. MHA通过脚本(masterha_master_monitor)实时(ping_interval=2)监控主库的状态(主机,数据库状态)
2. 监控到主库宕机之后,会发生自动选主(masterha_master_switch )
    (1) 权重: 参数
    (2) 日志量
    (3) 配置文件的顺序
3. 数据补偿
    (1) 当主库SSH能够连接,各个从库会立即保存(save_binary_logs)缺失部分的日志到/var/tmp,进行数据补偿.
    (2) 当主库SSH无法连接,通过脚本(apply_diff_relay_logs)自动进行从库的relaylog的差异处理,进行数据补偿.
4. 构建新的主从关系,并将故障节点移除(配置文件移除)    
5. MHA主进程完成工作后,自动退出.

6. 应用透明 VIP(master_ip_failover_script=/usr/local/bin/master_ip_failover)
7. 故障通知 (report_script=/usr/local/bin/send)
8. 提供额外数据补偿的手段(binlog server)

   自动自愈,待开发.... 

Architecture Description:

1主2从，

master：db01   

slave：db02   db03 ）：
MHA 高可用方案软件构成
Manager软件：选择一个从节点安装
Node软件：所有节点都要安装

MHA Software Configuration

Manager 工具包主要包括以下几个工具：

masterha_manger             # 启动MHA 
masterha_check_ssh          #检查MHA的SSH配置状况 
masterha_check_repl         #检查MySQL复制状况 
masterha_master_monitor     #检测master是否宕机 
masterha_check_status       #检测当前MHA运行状态 
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

Node工具包主要包括以下几个工具：
这些工具通常由MHA Manager的脚本触发，无需人为操作
save_binary_logs            保存和复制master的二进制日志 
apply_diff_relay_logs       识别差异的中继日志事件并将其差异的事件应用于其他的
purge_relay_logs            清除中继日志（不会阻塞SQL线程）

MHA environment to build

planning:

主库: 51    node 
从库: 
52      node
53      node    manager

Preparing the environment

(2 slightly from the main GTID .1)

Configuration program soft key connection

ln -s /application/mysql/bin/mysqlbinlog    /usr/bin/mysqlbinlog
ln -s /application/mysql/bin/mysql          /usr/bin/mysql

Configure each node trust

The first time you connect manually yes to confirm,

db01：

rm -rf /root/.ssh 
ssh-keygen
cd /root/.ssh 
mv id_rsa.pub authorized_keys
scp  -r  /root/.ssh  10.0.0.52:/root 
scp  -r  /root/.ssh  10.0.0.53:/root 

各节点验证
db01:
ssh 10.0.0.51 date
ssh 10.0.0.52 date
ssh 10.0.0.53 date

db02:
ssh 10.0.0.51 date
ssh 10.0.0.52 date
ssh 10.0.0.53 date

db03:
ssh 10.0.0.51 date
ssh 10.0.0.52 date
ssh 10.0.0.53 date

install software

Download software mha

mha official website: HTTPS: //code.google.com/archive/p/mysql-master-ha/
GitHub Download: https: //github.com/yoshinorim/mha4mysql-manager/wiki/Downloads

All nodes installed Node software dependencies

yum install perl-DBD-MySQL -y
rpm -ivh mha4mysql-node-0.56-0.el6.noarch.rpm

Mha need to create a user in the master database db01

grant all privileges on *.* to mha@'10.0.0.%' identified by 'mha';

Manager Software Installation (db03)

yum install -y perl-Config-Tiny epel-release perl-Log-Dispatch perl-Parallel-ForkManager perl-Time-HiRes
rpm -ivh mha4mysql-manager-0.56-0.el6.noarch.rpm

Configuration file preparation (db03)

Create a profile directory

mkdir -p /etc/mha

Create a log directory

mkdir -p /var/log/mha/app1

Edit mhaProfile

[root@db03 ~]#  vim /etc/mha/app1.cnf

[server default]
manager_log=/var/log/mha/app1/manager        
manager_workdir=/var/log/mha/app1            
master_binlog_dir=/data/binlog       
user=mha                                   
password=mha                               
ping_interval=2
repl_password=123
repl_user=repl
ssh_user=root                               
[server1]                                   
hostname=10.0.0.51
port=3306                                  
[server2]            
hostname=10.0.0.52
port=3306
[server3]
hostname=10.0.0.53
port=3306

Status Check

## 互信检查
[root@db03 ~]# masterha_check_ssh  --conf=/etc/mha/app1.cnf 


## 主从状态检查
[root@db03 ~]# masterha_check_repl  --conf=/etc/mha/app1.cnf 

MySQL Replication Health is OK.

Open MHA (db03):

[root@db03 ~]#  masterha_manager --conf=/etc/mha/app1.cnf --remove_dead_master_conf --ignore_last_failover  < /dev/null> /var/log/mha/app1/manager.log 2>&1 &

View MHA status

[root@db03 ~]# masterha_check_status --conf=/etc/mha/app1.cnf
app1 (pid:4719) is running(0:PING_OK), master:10.0.0.51

2.4.10 fault simulation and processing

(1) Stop the main library db01:

[root@db01 ~]# /etc/init.d/mysqld stop

观察manager  日志 tail -f /var/log/mha/app1/manager
末尾必须显示successfully，才算正常切换成功。                           

(2) repair the main library

[root@db01 ~]# /etc/init.d/mysqld start

(3) from the main repair

[root@db01 ~]# mysql -e "change master to master_host='10.0.0.52',master_user='repl',master_password='123' ,MASTER_AUTO_POSITION=1;"
[root@db01 ~]# mysql -e  "start slave;"
[root@db01 ~]# mysql -e  "show slave status \G"|grep Running:

Slave_IO_Running: Yes
Slave_SQL_Running: Yes

(4) restore configuration files

[root@db03 /etc/mha]# cat /etc/mha/app1.cnf 
[server default]
manager_log=/var/log/mha/app1/manager
manager_workdir=/var/log/mha/app1
master_binlog_dir=/data/binlog
password=mha
ping_interval=2
repl_password=123
repl_user=repl
ssh_user=root
user=mha
[server1]
hostname=10.0.0.51
port=3306

[server2]
hostname=10.0.0.52
port=3306

[server3]
hostname=10.0.0.53
port=3306 

(5) Start MHA

[root@db03 ~]# nohup masterha_manager --conf=/etc/mha/app1.cnf --remove_dead_master_conf --ignore_last_failover  < /dev/null> /var/log/mha/app1/manager.log 2>&1 &
    
[root@db03 ~]# masterha_check_status --conf=/etc/mha/app1.cnf

Manager introduces additional parameters

Description: The
main library is down someone to take over?

1. All logs from the node are the same, the default will be the order of the configuration file to select a new master.
2. Log inconsistency from the node, automatically selects the closest to the main library from the library
3. If the weight is set (candidate_master = 1), the weight for a preferred node node.
   But this amount of log node 100M log behind the main library, it would not be selected. May be blended check_repl_delay = 0, close inspection of the amount of the log, the forced selection candidate nodes.

(1)  ping_interval=1
# 设置监控主库，发送ping包的时间间隔，尝试三次没有回应的时候自动进行failover

(2) candidate_master=1
# 设置为候选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

(3)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

MHA function of vip

(1) preparation vip script

[root@db03 ~]# cp master_ip_failover.txt  /usr/local/bin/master_ip_failover
[root@db03 ~]# cd /usr/local/bin/
[root@db03 /usr/local/bin]# yum install -y dos2unix
[root@db03 /usr/local/bin]# dos2unix master_ip_failover 
[root@db03 /usr/local/bin]# chmod +x master_ip_failover 

(2) to change the manager configuration file:

vi /etc/mha/app1.cnf

#添加：
master_ip_failover_script=/usr/local/bin/master_ip_failover

(3) modify the contents of the file:

[root@db03 /usr/local/bin]# vim /usr/local/bin/master_ip_failover 

my $vip = '10.0.0.55/24';
my $key = '1';
my $ssh_start_vip = "/sbin/ifconfig eth0:$key $vip";
my $ssh_stop_vip = "/sbin/ifconfig eth0:$key down";

(4) on a main database, a manually generated first address vip

Hand bound to the main library vip, attention must be consistent ethN and configuration files, I was eth0: (1 is the key value specified) 1

ifconfig eth0:1 10.0.0.55/24

(5) Restart mha

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

nohup masterha_manager --conf=/etc/mha/app1.cnf --remove_dead_master_conf --ignore_last_failover < /dev/null > /var/log/mha/app1/manager.log 2>&1 &

email alert

(1) add configuration options

vi /etc/mha/app1.cnf
report_script=/usr/local/bin/send

(2) a copy of the script

[root@db03 ~]# cp -a email/*  /usr/local/bin

(3) Stop MHA

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

(4) open MHA

nohup masterha_manager --conf=/etc/mha/app1.cnf --remove_dead_master_conf --ignore_last_failover < /dev/null > /var/log/mha/app1/manager.log 2>&1 &

(5) close the main library, see warning messages

After the main library is closed, it will automatically switch the main library at the same time send a warning message. But the MHA will stop, we need to re-build MHA

binlog server（db03）

(1) add configuration information

vim /etc/mha/app1.cnf 
[binlog1]
no_master=1
hostname=10.0.0.53
master_binlog_dir=/data/mysql/binlog

(2) create the necessary directories

mkdir -p /data/mysql/binlog
chown -R mysql.mysql /data/*

(3) pulls the master database log binlog

Note: The starting point for pulling logs, according to the current needs have been acquired from the binary log starting point library

cd /data/mysql/binlog
mysqlbinlog  -R --host=10.0.0.51 --user=mha --password=mha --raw  --stop-never mysql-bin.000001 &

(4) Restart MHA

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

nohup masterha_manager --conf=/etc/mha/app1.cnf --remove_dead_master_conf --ignore_last_failover < /dev/null > /var/log/mha/app1/manager.log 2>&1 &

MHA Troubleshooting

Check the status of MHA

masterha_check_status --conf=/etc/mha/app1.cnf

Check the configuration file mha

[root@db03 ~]# vim /etc/mha/app1.cnf 
[server default]
manager_log=/var/log/mha/app1/manager
manager_workdir=/var/log/mha/app1
master_binlog_dir=/data/binlog
master_ip_failover_script=/usr/local/bin/master_ip_failover
password=mha
ping_interval=2
repl_password=123
repl_user=repl
report_script=/usr/local/bin/send
ssh_user=root
user=mha

[server1]
hostname=10.0.0.51
port=3306

[server2]
hostname=10.0.0.52
port=3306

[server3]
hostname=10.0.0.53
port=3306

[binlog1]
no_master=1
hostname=10.0.0.53
master_binlog_dir=/data/mysql/binlog

3. Check the relationship of mutual trust

[root@db03 ~]# masterha_check_ssh --conf=/etc/mha/app1.cnf

4. Check the script

At the same time need to check the contents of the script is correct

ls -l /usr/local/bin/
-rwxr-xr-x 1 root root  2166 Aug 20 15:45 master_ip_failover
-rwxr-xr-x 1 root root    35 Aug 20 15:51 send
-rwxr-xr-x 1 root root 80213 Aug 20 15:51 sendEmail
-rwxr-xr-x 1 root root   198 Aug 20 16:01 testpl

5. Check binlogserver

ps -ef| grep mysqlbinlog

6. Make sure that the main library, the recovery binlog log

db02 [(none)]>show master status


cd /data/mysql/binlog
rm -rf *
mysqlbinlog -R --host=10.0.0.52 --user=mha --password=mha --raw --stop-never mysql-bin.000005 &

7. Check the main library VIP

ifconfig eth0:1 10.0.0.55/24
ifconfig

8. Check the status of master and slave

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

9. Start MHA

nohup masterha_manager --conf=/etc/mha/app1.cnf --remove_dead_master_conf --ignore_last_failover < /dev/null > /var/log/mha/app1/manager.log 2>&1 &

masterha_check_status --conf=/etc/mha/app1.cnf