如何在多个数据中心部署多套MGR集群，并实现故障快速切换。

上篇文章介绍了如何在多数据中心部署多套MGR集群，并构建集群间的复制通道。这样一旦主AZ不可用时，在校验完数据后，就可以切换到备用AZ的MGR集群，非常方便。

本文我们继续深入介绍如何利用 Async Replication Auto failover 实现故障自动转移的。

1、什么是Async Replication Auto failover

从MySQL 8.0.22开始，推出一个新特性"Async Replication Auto failover"，当MGR集群发生故障时，其从库可以更方便的实现快速自动切主。直译过来是“异步复制自动故障转移”，但实际上它也是支持半同步复制场景的。

You can use MySQL Server's new asynchronous connection failover mechanism to automatically establish an asynchronous (source to replica) replication connection to a new source after the existing connection from a replica to its source fails. The connection fails over if the replication I/O thread stops due to the source stopping or due to a network failure. The asynchronous connection failover mechanism can be used to keep a replica synchronized with multiple MySQL servers or groups of servers that share data. To activate asynchronous connection failover for a replication channel set SOURCE_CONNECTION_AUTO_FAILOVER=1 on the CHANGE MASTER TO statement for the channel, and set up a source list for the channel using the asynchronous_connection_failover_add_source and asynchronous_connection_failover_delete_source functions.

详细介绍见官方文档 17.4.9 Switching Sources with Asynchronous Connection Failover

2、基于MGR的两地三中心数据库架构方案

在两地三中心架构下，可以采用下面这个部署方案

在这个架构方案里，MGR-B可以采用 异步复制 或 增强半同步复制 通道从MGR-A复制数据，这要取决于两个AZ之间的网络状况。

在金融应用场景下，这个网络条件一般可以得到保障，因此优先采用增强版同步方式。

而跨城异地AZ里的MGR C则因为网络延迟较大，大概率会采用异步复制方式。

在上述方案中，不管是MGR-B还是C，都面临一个问题：那就是复制源指向的主机实例，发生故障不可用之后，如何快速切换，实现自动故障转移。

在以往，只能靠第三方工具实现切换。

在MySQL 8.0.22新增"Async Replication Auto failover"特性后，就没这个烦恼了。

其工作机制是 在一个复制通道上设置多个复制源（source），它还支持对多个源设置不同权重。当发现主复制源发生故障异常中断后（会先尝试重连几次），即可实现自动切换到新的复制源。当原来的复制源恢复后，如果其权重更高，还会再切换回去。

3、配置Async Replication Auto failover

部署的过程很简单，几条命令就搞定了。

3.1、创建复制通道

按照常规方式，在从实例上（本案以MGR-B为例）创建一个复制通道

[root@GreatSQL mgrB-1][(none)]> CHANGE REPLICATION SOURCE TO
MASTER_HOST='172.16.16.10', MASTER_PORT=3306, 
MASTER_USER='repl', MASTER_PASSWORD='repl',
MASTER_AUTO_POSITION=1,
SOURCE_CONNECTION_AUTO_FAILOVER=1, #这里是关键，表示开启自动故障转移
MASTER_RETRY_COUNT=3, #最多重试3次
MASTER_CONNECT_RETRY=10 #每次重试间隔10秒
FOR CHANNEL 'MGR-A';

#简单解释下几个参数
- SOURCE_CONNECTION_AUTO_FAILOVER=1  #这里是关键，表示开启自动故障转移
- MASTER_RETRY_COUNT=3  #表示最多重试3次，默认是是86400次
- MASTER_CONNECT_RETRY=10 #表示每次重试间隔10秒，默认是60秒

确认添加的复制通道生效了：

[root@GreatSQL mgrB-1][(none)]> SELECT * FROM performance_schema.replication_applier_status\G
*************************** 1. row ***************************
              CHANNEL_NAME: mgr-a
             SERVICE_STATE: ON
           REMAINING_DELAY: NULL
COUNT_TRANSACTIONS_RETRIES: 0
*************************** 2. row ***************************
              CHANNEL_NAME: group_replication_applier
             SERVICE_STATE: ON
           REMAINING_DELAY: NULL
COUNT_TRANSACTIONS_RETRIES: 0

3.2、对复制通道添加多个复制源

接下来再对这个复制通道添加多个复制源（多次调用该UDF即可）：

[root@GreatSQL mgrB-1][(none)]> SELECT asynchronous_connection_failover_add_source('MGR-A','172.16.16.10',3306,null,60);
[root@GreatSQL mgrB-1][(none)]> SELECT asynchronous_connection_failover_add_source('MGR-A','172.16.16.11',3306,null,60);
[root@GreatSQL mgrB-1][(none)]> SELECT asynchronous_connection_failover_add_source('MGR-A','172.16.16.12',3306,null,60);

简单解释下几个参数

• MGR-A #表示复制通道，和上面的复制通道同名
• 172.16.16.10 #表示该复制源的IP
• 3306 #表示该复制源的端口
• null #表示network_namespace，未来的特性，现在先放空即可
• 60 #表示该复制源的权重，上面我们介绍了不同权重的作用，值越大越有机会抢到成为复制源

确认多个复制源生效：

[root@GreatSQL mgrB-1][(none)]> SELECT * FROM performance_schema.replication_asynchronous_connection_failover;
+--------------+--------------+------+-------------------+--------+--------------+
| CHANNEL_NAME | HOST         | PORT | NETWORK_NAMESPACE | WEIGHT | MANAGED_NAME |
+--------------+--------------+------+-------------------+--------+--------------+
| mgr-a        | 172.16.16.10 | 3306 |                   |     60 |              |
| mgr-a        | 172.16.16.11 | 3306 |                   |     60 |              |
| mgr-a        | 172.16.16.12 | 3306 |                   |     60 |              |
+--------------+--------------+------+-------------------+--------+--------------+

启动该复制通道：

[root@GreatSQL mgrB-1][(none)]> START REPLICA FOR CHANNEL 'MGR-A';

确认复制通道和MGR的状态都正常：

[root@GreatSQL mgrB-1][(none)]> SELECT * FROM performance_schema.replication_connection_status\G
*************************** 1. row ***************************
                                      CHANNEL_NAME: mgr-a
                                        GROUP_NAME:
                                       SOURCE_UUID: b084f8a1-96a8-11eb-9a70-525400fb993a
                                         THREAD_ID: 3084
                                     SERVICE_STATE: ON
                         COUNT_RECEIVED_HEARTBEATS: 5974
                          LAST_HEARTBEAT_TIMESTAMP: 2021-05-29 18:53:13.879720
                          RECEIVED_TRANSACTION_SET: 476c0276-be03-11eb-bd34-525400e802e2:21-31:1000016-1000017
                                 LAST_ERROR_NUMBER: 0
                                LAST_ERROR_MESSAGE:
                              LAST_ERROR_TIMESTAMP: 0000-00-00 00:00:00.000000
                           LAST_QUEUED_TRANSACTION: 476c0276-be03-11eb-bd34-525400e802e2:31
 LAST_QUEUED_TRANSACTION_ORIGINAL_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
LAST_QUEUED_TRANSACTION_IMMEDIATE_COMMIT_TIMESTAMP: 2021-05-27 17:19:43.201000
     LAST_QUEUED_TRANSACTION_START_QUEUE_TIMESTAMP: 2021-05-27 17:19:43.203315
       LAST_QUEUED_TRANSACTION_END_QUEUE_TIMESTAMP: 2021-05-27 17:19:43.203349
                              QUEUEING_TRANSACTION:
    QUEUEING_TRANSACTION_ORIGINAL_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
   QUEUEING_TRANSACTION_IMMEDIATE_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
        QUEUEING_TRANSACTION_START_QUEUE_TIMESTAMP: 0000-00-00 00:00:00.000000
*************************** 2. row ***************************
                                      CHANNEL_NAME: group_replication_applier
                                        GROUP_NAME: f195537d-19ac-11eb-b29f-5254002eb6d6
                                       SOURCE_UUID: f195537d-19ac-11eb-b29f-5254002eb6d6
                                         THREAD_ID: NULL
                                     SERVICE_STATE: ON
                         COUNT_RECEIVED_HEARTBEATS: 0
                          LAST_HEARTBEAT_TIMESTAMP: 0000-00-00 00:00:00.000000
                          RECEIVED_TRANSACTION_SET: 476c0276-be03-11eb-bd34-525400e802e2:1-31:1000015-1000017,
f195537d-19ac-11eb-b29f-5254002eb6d6:1-18
                                 LAST_ERROR_NUMBER: 0
                                LAST_ERROR_MESSAGE:
                              LAST_ERROR_TIMESTAMP: 0000-00-00 00:00:00.000000
                           LAST_QUEUED_TRANSACTION: f195537d-19ac-11eb-b29f-5254002eb6d6:18
 LAST_QUEUED_TRANSACTION_ORIGINAL_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
LAST_QUEUED_TRANSACTION_IMMEDIATE_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
     LAST_QUEUED_TRANSACTION_START_QUEUE_TIMESTAMP: 2021-05-27 17:04:03.407281
       LAST_QUEUED_TRANSACTION_END_QUEUE_TIMESTAMP: 2021-05-27 17:04:03.407317
                              QUEUEING_TRANSACTION:
    QUEUEING_TRANSACTION_ORIGINAL_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
   QUEUEING_TRANSACTION_IMMEDIATE_COMMIT_TIMESTAMP: 0000-00-00 00:00:00.000000
        QUEUEING_TRANSACTION_START_QUEUE_TIMESTAMP: 0000-00-00 00:00:00.000000

执行 SHOW REPLICA STATUS 查看状态：

*************************** 1. row ***************************
             Replica_IO_State: Waiting for master to send event
                  Source_Host: 172.16.16.10
                  Source_User: repl
                  Source_Port: 3306
                Connect_Retry: 10
...
           Replica_IO_Running: Yes
          Replica_SQL_Running: Yes
...
                  Source_UUID: 5499a6cb-91cb-11eb-966f-525400e802e2
...
           Source_Retry_Count: 3
...
           Retrieved_Gtid_Set: 476c0276-be03-11eb-bd34-525400e802e2:21-31:1000016-1000017
            Executed_Gtid_Set: 476c0276-be03-11eb-bd34-525400e802e2:1-31:1000015-1000017,
f195537d-19ac-11eb-b29f-5254002eb6d6:1-18
                Auto_Position: 1
         Replicate_Rewrite_DB:
                 Channel_Name: mgr-a
...               

先记住上面输出结果中的 Source_Host 和 Source_UUID 等信息，下面模拟一次复制源服务器宕机后，自动切换复制源的场景。

4、模拟故障，确认可自动切换

在当前复制源服务器上，执行 kill -9 杀掉 mysqld 进程，然后就能看到从服务器上有类似如下日志：

# 先尝试3次（每次间隔10秒）重连旧的复制源服务器
[ERROR] [MY-010584] [Repl] Slave I/O for channel 'mgr-a': error connecting to master 'repl@172.16.16.10:3306' - retry-time: 10 retries: 3 message: Can't connect to MySQL server on '172.16.16.10:3306' (111), Error_code: MY-002003

#重试失败，停止复制I/O线程
[Note] [MY-010563] [Repl] Slave I/O thread for channel 'mgr-a' killed while connecting to master
[Warning] [MY-010897] [Repl] Storing MySQL user name or password information in the master info repository is not secure and is therefore not recommended. Please consider using the USER and PASSWORD connection options for START SLAVE; see the 'START SLAVE Syntax' in the MySQL Manual for more information.

# 再次启动复制I/O线程，连接到新的复制源服务器
[System] [MY-010562] [Repl] Slave I/O thread for channel 'mgr-a': connected to master 'repl@172.16.16.11:3306',replication started in log 'FIRST' at position 8598
# 告知UUID发生切换了
[Warning] [MY-010549] [Repl] The master's UUID has changed, although this should not happen unless you have changed it manually. The old UUID was ec2fcbeb-976c-11eb-a652-525400e2078a.

再次执行 SHOW REPLICA STATUS 确认复制源切换了：

*************************** 1. row ***************************
             Replica_IO_State: Waiting for master to send event
                  Source_Host: 172.16.16.11
                  Source_User: repl
                  Source_Port: 3306
                Connect_Retry: 10
...
           Replica_IO_Running: Yes
          Replica_SQL_Running: Yes
...
                  Source_UUID: ec2fcbeb-976c-11eb-a652-525400e2078a
...
           Source_Retry_Count: 3
...
           Retrieved_Gtid_Set: 476c0276-be03-11eb-bd34-525400e802e2:21-32:1000016-1000017
            Executed_Gtid_Set: 476c0276-be03-11eb-bd34-525400e802e2:1-32:1000015-1000017,
f195537d-19ac-11eb-b29f-5254002eb6d6:1-18
                Auto_Position: 1
         Replicate_Rewrite_DB:
                 Channel_Name: mgr-a
...

因为3个复制源的权重设置为一样，所以当原来的复制源服务器宕机恢复后，不会再切换回去。而如果旧的复制源服务器权重设置较高的话，当他恢复后，会再次发生切换，切回原来的源：

#没有任何尝试重连的行为，直接发起切换
[Note] [MY-011026] [Repl] Slave I/O thread killed while reading event for channel 'mgr-a'.
[Note] [MY-010570] [Repl] Slave I/O thread exiting for channel 'mgr-a', read up to log 'FIRST', position 8871
[Warning] [MY-010897] [Repl] Storing MySQL user name or password information in the master info repository is not secure and is therefore not recommended. Please consider using the USER and PASSWORD connection options for START SLAVE; see the 'START SLAVE Syntax' in the MySQL Manual for more information.
[System] [MY-010562] [Repl] Slave I/O thread for channel 'mgr-a': connected to master 'repl@172.16.16.10:3306',replication started in log 'FIRST' at position 8871
[Warning] [MY-010549] [Repl] The master's UUID has changed, although this should not happen unless you have changed it manually. The old UUID was 5499a6cb-91cb-11eb-966f-525400e802e2. -- 再次切回原来的主

这就很方便的可以实现自动故障转移了。

现在，我们利用MGR + 增强半同步复制 + 自动故障转移 构建了一套金融级应用场景下的两地多中心数据库架构方案。推荐选用可靠性、稳定性更高的GreatSQL，可以更放心的使用MGR（GreatSQL，打造更好的MGR生态）。

后面再继续介绍基于MGR的其他架构解决方案。

Enjoy GreatSQL :)