1,关于MySQL Group Replication
基于组的复制(Group-basedReplication)是一种被使用在容错系统中的技术。Replication-group(复制组)是由能够相互通信的多个服务器(节点)组成的。
在通信层,Groupreplication实现了一系列的机制:比如原子消息(atomicmessage delivery)和全序化消息(totalorderingof messages)。
这些原子化,抽象化的机制,为实现更先进的数据库复制方案提供了强有力的支持。
MySQL Group Replication正是基于这些技术和概念,实现了一种多主全更新的复制协议。
简而言之,一个Replication-group就是一组节点,每个节点都可以独立执行事务,而读写事务则会在于group内的其他节点进行协调之后再commit。
因此,当一个事务准备提交时,会自动在group内进行原子性的广播,告知其他节点变更了什么内容/执行了什么事务。
这种原子广播的方式,使得这个事务在每一个节点上都保持着同样顺序。
这意味着每一个节点都以同样的顺序,接收到了同样的事务日志,所以每一个节点以同样的顺序重演了这些事务日志,最终整个group保持了完全一致的状态。
然而,不同的节点上执行的事务之间有可能存在资源争用。这种现象容易出现在两个不同的并发事务上。
假设在不同的节点上有两个并发事务,更新了同一行数据,那么就会发生资源争用。
面对这种情况,GroupReplication判定先提交的事务为有效事务,会在整个group里面重演,后提交的事务会直接中断,或者回滚,最后丢弃掉。
因此,这也是一个无共享的复制方案,每一个节点都保存了完整的数据副本。看如下图片01.png,描述了具体的工作流程,能够简洁的和其他方案进行对比。这个复制方案,在某种程度上,和数据库状态机(DBSM)的Replication方法比较类似。
MGR的限制
仅支持InnoDB表,并且每张表一定要有一个主键,用于做write set的冲突检测;
必须打开GTID特性,二进制日志格式必须设置为ROW,用于选主与write set
COMMIT可能会导致失败,类似于快照事务隔离级别的失败场景
目前一个MGR集群最多支持9个节点
不支持外键于save point特性,无法做全局间的约束检测与部分部分回滚
二进制日志不支持binlog event checksum
2,安装mysql 5.7.18
在三台db服务器上面设置/etc/hosts映射,如下:
192.168.1.20 mgr1
192.168.1.21 mgr2
192.168.1.22 mgr3
安装的数据库服务器:
数据库服务器地址 | 端口 | 数据目录 | Server-id |
192.168.1.20(mgr1) | 3306 | /usr/local/mysql/data | 20 |
192.168.1.21(mgr2) | 3306 | /usr/local/mysql/data | 21 |
192.168.1.22(mgr3) | 3306 | /usr/local/mysql/data | 22 |
安装过程略。。。。。
my.cnf的详细配置
[client]
port =
socket = /usr/local/mysql/tmp/mysql.sock [mysqld]
port =
socket = /usr/local/mysql/tmp/mysql.sock
back_log =
basedir = /usr/local/mysql
tmpdir = /tmp
datadir = /usr/local/mysql/data #-------------------gobal variables------------#
gtid_mode = ON
enforce_gtid_consistency = ON
master_info_repository = TABLE
relay_log_info_repository = TABLE
binlog_checksum = NONE
log_slave_updates = ON
log-bin = /usr/local/mysql/log/mysql-bin
transaction_write_set_extraction = XXHASH64
loose-group_replication_group_name = 'ce9be252-2b71-11e6-b8f4-00212844f856'
loose-group_replication_start_on_boot = off
loose-group_replication_local_address = '192.168.1.20:33061'
loose-group_replication_group_seeds ='192.168.1.20:33061,192.168.1.21:33061,192.168.1.22:33061'
loose-group_replication_bootstrap_group = off
max_connect_errors =
max_connections =
wait_timeout =
interactive_timeout =
net_read_timeout =
net_write_timeout =
table_open_cache =
table_definition_cache =
thread_cache_size =
open_files_limit =
character-set-server = utf8
collation-server = utf8_bin
skip_external_locking
performance_schema =
user = mysql
myisam_recover_options = DEFAULT
skip-name-resolve
local_infile =
lower_case_table_names = #--------------------innoDB------------#
innodb_buffer_pool_size = 2000M
innodb_data_file_path = ibdata1:200M:autoextend
innodb_flush_log_at_trx_commit =
innodb_io_capacity =
innodb_lock_wait_timeout =
innodb_log_buffer_size = 8M
innodb_log_file_size = 200M
innodb_log_files_in_group =
innodb_max_dirty_pages_pct =
innodb_read_io_threads =
innodb_write_io_threads =
innodb_support_xa =
innodb_thread_concurrency =
innodb_file_per_table
innodb_rollback_on_timeout #------------session variables-------#
join_buffer_size = 8M
key_buffer_size = 256M
bulk_insert_buffer_size = 8M
max_heap_table_size = 96M
tmp_table_size = 96M
read_buffer_size = 8M
sort_buffer_size = 2M
max_allowed_packet = 64M
read_rnd_buffer_size = 32M #------------MySQL Log----------------#
log-bin = my3306-bin
binlog_format = row
sync_binlog =
expire_logs_days =
max_binlog_cache_size = 128M
max_binlog_size = 500M
binlog_cache_size = 64k
slow_query_log
log-slow-admin-statements
log_warnings =
long_query_time = 0.25 #---------------replicate--------------#
relay-log-index = relay3306.index
relay-log = relay3306
server-id =
init_slave = 'set sql_mode=STRICT_ALL_TABLES'
log-slave-updates
[myisamchk]
key_buffer = 512M
sort_buffer_size = 512M
read_buffer = 8M
write_buffer = 8M
[mysqlhotcopy]
interactive-timeout
[mysqld_safe]
open-files-limit =
log-error = /usr/local/mysql/log/mysqld_error.log
my.cnf
注意每个节点的server_id、loose-group_replication_local_address、loose-group_replication_group_seeds都配置成自己的相印的参数
3,创建复制环境
在mgr1/mgr2/mgr3上建立复制账号:
mysql> GRANT REPLICATION SLAVE ON *.* TO 'repl'@'192.168.%' IDENTIFIED BY 'repl';
4,安装group replication插件
在mgr1、mgr2、mgr3上依次安装group replication插件
mysql> INSTALL PLUGIN group_replication SONAME 'group_replication.so';
5,配置group replication参数
确保binlog_format是row格式。
mysql> show variables like 'binlog_format';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| binlog_format | ROW |
+---------------+-------+
1 row in set (0.00 sec)
配置文件配置
(1) mgr1上的my.cnf配置:
server-id=20
transaction_write_set_extraction = XXHASH64
loose-group_replication_group_name = "ce9be252-2b71-11e6-b8f4-00212844f856"
loose-group_replication_start_on_boot = off
loose-group_replication_local_address = "192.168.1.20:33061"
loose-group_replication_group_seeds = "192.168.1.20:33061,192.168.1.21:33061,192.168.1.22:33061"
loose-group_replication_bootstrap_group = off
loose-group_replication_single_primary_mode = true
loose-group_replication_enforce_update_everywhere_checks = false (2)mgr2上的my.cnf配置: server-id=21
transaction_write_set_extraction = XXHASH64
loose-group_replication_group_name = "ce9be252-2b71-11e6-b8f4-00212844f856"
loose-group_replication_start_on_boot = off
loose-group_replication_local_address = "192.168.1.21:33061"
loose-group_replication_group_seeds = "192.168.1.20:33061,192.168.1.21:33061,192.168.1.22:33061"
loose-group_replication_bootstrap_group = off
loose-group_replication_single_primary_mode = true
loose-group_replication_enforce_update_everywhere_checks = false (3)mgr3上的my.cnf配置: server-id=22
transaction_write_set_extraction = XXHASH64
loose-group_replication_group_name = "ce9be252-2b71-11e6-b8f4-00212844f856"
loose-group_replication_start_on_boot = off
loose-group_replication_local_address = "192.168.1.22 :33061"
loose-group_replication_group_seeds = "192.168.1.20:33061,192.168.1.21:33061,192.168.1.22:33061"
loose-group_replication_bootstrap_group = off
loose-group_replication_single_primary_mode = true
loose-group_replication_enforce_update_everywhere_checks = false |
配置完后,重启3个db上的mysql服务
6,启动mgr集群
开始构建group replication集群,通常操作命令
在mgr1、mgr2、mgr3上依次执行
mysql> CHANGE MASTER TO MASTER_USER='repl', MASTER_PASSWORD='repl' FOR CHANNEL 'group_replication_recovery';
Query OK, 0 rows affected, 2 warnings (0.02 sec)
Db1上建立基本主库master库:
# 设置group_replication_bootstrap_group为ON是为了标示以后加入集群的服务器以这台服务器为基准,以后加入的就不需要设置。
mysql> SET GLOBAL group_replication_bootstrap_group = ON;
Query OK, 0 rows affected (0.00 sec)
mysql> START GROUP_REPLICATION;
Query OK, 0 rows affected (1.03 sec)
mysql> select * from performance_schema.replication_group_members;
Db2上启动group_replication:
Db2上mysql命令行上执行启动:
mysql> set global group_replication_allow_local_disjoint_gtids_join=ON;
mysql> start group_replication;
mysql> select * from performance_schema.replication_group_members;
Db3上启动group_replication:
-- Db3命令行上执行:
mysql> set global group_replication_allow_local_disjoint_gtids_join=ON;
mysql> start group_replication;
-- 再去master库mgr1上,查看group_replication成员,会有mgr3的显示,而且已经是ONLINE了
mysql> select * from performance_schema.replication_group_members;
最后查看集群状态,都为ONLINE就表示OK:
mysql> select * from performance_schema.replication_group_members;
7,验证集群复制功能
查看那个是主节点
mysql> select variable_value from performance_schema.global_status where variable_name="group_replication_primary_member";
在master库mgr1上建立测试库mgr1,测试表t1,录入一条数据
mysql> create database mgr1;
Query OK, 1 row affected (0.00 sec)
mysql> create table mgr1.t1(id int,cn varchar(32));
Query OK, 0 rows affected (0.02 sec)
mysql> insert into t1(id,cn)values(1,'a');
ERROR 3098 (HY000): The table does notcomply with the requirements by an external plugin.
-- # 这里原因是group_replaction环境下面,表必须有主键不然不允许往里insert值。所以修改表t1,将id字段设置程主键即可。
mysql> alter table t1 add primary key(id);
mysql> insert into t1(id,cn)values(1,'a');
去mgr2/mgr3上可以看到数据已经同步过去
mysql> select * from mgr1.t1;
+----+------+
| id | cn |
+----+------+
| 1| a |
+----+------+
1 row in set (0.00 sec)
然后在mgr2/mgr3上执行inert操作,则拒绝,因为mgr2、mgr3为readonly
mysql> insert into t1 select 2,'b';
ERROR 1290 (HY000): The MySQL server isrunning with the --super-read-only option so it cannot execute this statement
8,问题记录
8.1问题记录一
MySQL窗口报错:
ERROR 3092 (HY000): The server is notconfigured properly to be an active member of the group. Please see moredetails on error log.
后台ERROR LOG报错:
[ERROR] Plugin group_replication reported:'This member has more executed transactions than those present in the group.Local transactions: f16f7f74-c283-11e6-ae37-fa163ee40410:1 > Grouptransactions: 3c992270-c282-11e6-93bf-fa163ee40410:1,
aaaaaa:1-5'
[ERROR]Plugin group_replication reported: 'The member contains transactions notpresent in the group. The member will now exit the group.'
[Note] Plugin group_replication reported: 'Toforce this member into the group you can use the group_replication_allow_local_disjoint_gtids_joinoption'
【解决办法】:
根据提示打开group_replication_allow_local_disjoint_gtids_join选项,mysql命令行执行:
mysql> set globalgroup_replication_allow_local_disjoint_gtids_join=ON;
再执行开启组复制:
mysql> start group_replication;
Query OK, 0 rows affected (7.89 sec)
8.2 问题记录二RECOVERING
在mgr1上查询集群组成员
mysql> SELECT * FROM performance_schema.replication_group_members;
+---------------------------+--------------------------------------+----------------------+-------------+--------------+
| CHANNEL_NAME | MEMBER_ID | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE |
+---------------------------+--------------------------------------+----------------------+-------------+--------------+
| group_replication_applier | 3d872c2e-d670-11e6-ac1f-b8ca3af6e36c | | 3306 | ONLINE
| group_replication_applier | ef8ac2de-d671-11e6-9ba4-18a99b763071 | | 3306 | RECOVERING |
| group_replication_applier | fdf2b02e-d66f-11e6-98a8-18a99b76310d | | 3306 | RECOVERING |
+---------------------------+--------------------------------------+----------------------+-------------+--------------+
3 rows in set (0.00 sec)
【解决办法】:
看报错[ERROR] Slave I/O for channel'group_replication_recovery': error connecting to master 'repl@:3306'- retry-time: 60 retries: 1, Error_code:2005,连接master库不上,所以问题在这里,我们赋予的复制账号是ip的repl@'192.168.%',所以还需要做一个hostname()和mgr1的ip地址192.168.1.20的映射关系。
建立hostname和ip映射
vim /etc/hosts
192.168.1.20 mgr1
192.168.1.21 mgr2
192.168.1.22 mgr3
然后在mgr2上执行如下命令后重新开启group_replication即可。
mysql> stop group_replication;
Query OK, 0 rows affected (0.02 sec)
mysql> start group_replication;
Query OK, 0 rows affected (5.68 sec)
再去master库mgr1上,查看group_replication成员,会有mgr2的显示
mysql> SELECT * FROM performance_schema.replication_group_members;
+---------------------------+--------------------------------------+----------------------+-------------+--------------+
| CHANNEL_NAME | MEMBER_ID | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE
+---------------------------+--------------------------------------+----------------------+-------------+--------------+
| group_replication_applier | 3d872c2e-d670-11e6-ac1f-b8ca3af6e36c | | 3306 | ONLINE
| group_replication_applier | fdf2b02e-d66f-11e6-98a8-18a99b76310d | | 3306 | ONLINE |
+---------------------------+--------------------------------------+----------------------+-------------+--------------+
2 rows in set (0.00 sec)
8.3问题记录三
操作问题
mysql> START GROUP_REPLICATION;
ERROR 3092 (HY000): The server is notconfigured properly to be an active member of the group. Please see moredetails on error log.
【解决办法】:
mysql> SET GLOBALgroup_replication_bootstrap_group = ON;
mysql> START GROUP_REPLICATION;
9,日常维护步骤:
1、如果从库某一节点关闭
stop group_replication;
2、如果所有的库都关闭后,第一个库作为主库首先执行
set global group_replication_bootstrap_group=ON;
start group_replication;
剩下的库直接执行即可!
start group_replication;
3、如果主库故障,会自动从两个从库选出一个主库,主库启动后再次执行如下命令后会变成从库
start group_replication;