MySQL:slave_skip_errors参数对MGR可用性的影响


整个问题提出和查验由 @gc @甘露寺的姑子@乙酉 完毕,文档记载由@gc @乙酉完毕。

我仅仅进行了问题剖析和文档收拾


欢迎重视我的《深化了解MySQL主从原理 32讲 》,如下:

MySQL:slave_skip_errors参数对MGR可用性的影响

一、事例描绘

MGR在遇到表不存在的状况下,节点没有退出节点而是爆出一个正告,而且节点状况也正常,正告如下:

2019-10-17T21:16:11.564211+08:00 10 [Warning] Slave SQL for channel
group_replication_applier': Worker 1 failed executing transaction 'aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa:8' at master log , end_log_pos 220;
Error executing row event: 'Table 'test.a_1' doesn't exist', Error_code: 1146

集群状况如下:

[root@mysql.sock][test]>select * from performance_schema.replication_group_members;
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| CHANNEL_NAME              | MEMBER_ID                            | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| group_replication_applier | 9fd479bb-f0d8-11e9-9381-000c29105312 | mysql_1     |        3306 | ONLINE       |
| group_replication_applier | a8833a96-f0d8-11e9-a9f4-000c291fd9a5 | mysql_2     |        3306 | ONLINE       |
| group_replication_applier | b2968fe2-f0d8-11e9-a8ff-000c29c89e42 | mysql_3     |        3306 | ONLINE       |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
3 rows in set (0.00 sec)

其时觉得很古怪,咱们咱们都知道这种过失即便是在主从状况下也是报错的SQL线程退出的,MGR竟然还能在线,这样的一种状况数据现已不同步了,应该报错而且除掉节点才对。

二、问题剖析

随即一些感兴趣的同学立刻进行了查验,查验作用和上面不一致,查验作用是报错而不是出正告如下:

2019-10-17T09:16:34.317542Z 84 [ERROR] Slave SQL for channel
'group_replication_applier': Error executing row event:
'Table 'test.emp1' doesn't exist', Error_code: 1146

而且这样的一种状况表不存在的节点现已被除掉掉了。下面是一般的状况的节点状况:

secondary 1节点:
[root@mysql.sock][test]>select * from performance_schema.replication_group_members;
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| CHANNEL_NAME              | MEMBER_ID                            | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| group_replication_applier | a8833a96-f0d8-11e9-a9f4-000c291fd9a5 | mysql_2     |        3306 | ERROR        |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
1 row in set (0.00 sec)
 
secondary 2节点:
[root@mysql.sock][test]>select * from performance_schema.replication_group_members;
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| CHANNEL_NAME              | MEMBER_ID                            | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| group_replication_applier | b2968fe2-f0d8-11e9-a8ff-000c29c89e42 | mysql_3     |        3306 | ERROR        |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
1 row in set (0.00 sec)

那么疑问便是为什么同样是MGR一个是正告一个是过失呢,而且前者还能处于正常同步状况。不错看到标题就知道这儿和slave_skip_errors参数有关。

三、查验模仿

咱们咱们都知道再Master-Slave中假定遇到从库表不存在肯定是报错的,除非设置slave_skip_errors参数,当然我在线上重来没有设置过这个参数,而且经过这个事例咱们发现本参数对MGR也有影响,如下查验办法:

咱们在3个节点都翻开slave-skip-errors= ddl_exist_errors

如下图:

MySQL:slave_skip_errors参数对MGR可用性的影响

然后建立3节点single-primary方法的MGR集群

MySQL:slave_skip_errors参数对MGR可用性的影响

集群建立正常。

然后实施如下操作:

[root@mysql.sock][(none)]>set sql_log_bin=0;
Query OK, 0 rows affected (0.00 sec)
[root@mysql.sock][(none)]>create table test.a_1(id bigint auto_increment primary key,name varchar(20));
Query OK, 0 rows affected (0.01 sec)
[root@mysql.sock][(none)]>set sql_log_bin=1;
Query OK, 0 rows affected (0.00 sec)

此刻primary节点是有a_1表的,可是由于binlog封闭的原因,两个secondary节点是不存在a_1表的。

然后咱们刺进数据:

[root@mysql.sock][test]>insert into test.a_1 values(null,'tom');
Query OK, 1 row affected (0.02 sec)

此刻,primary节点由于存在a_1表,所以可以刺进,可是两个secondary节点不存在a_1表,所以刺进是失利的。数据发生不一致。一般的状况下这种数据不一致会导致2个secondary节点被提出集群才对。可是实际上3个节点都是正常的,集群并没有失效。

[root@mysql.sock][test]>select * from test.a_1;
+----+------+
| id | name |
+----+------+
|  1 | tom  |
+----+------+
1 row in set (0.00 sec)
[root@mysql.sock][test]>select * from performance_schema.replication_group_members;
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| CHANNEL_NAME              | MEMBER_ID                            | MEMBER_HOST | MEMBER_PORT | MEMBER_STATE |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
| group_replication_applier | 9fd479bb-f0d8-11e9-9381-000c29105312 | mysql_1     |        3306 | ONLINE       |
| group_replication_applier | a8833a96-f0d8-11e9-a9f4-000c291fd9a5 | mysql_2     |        3306 | ONLINE       |
| group_replication_applier | b2968fe2-f0d8-11e9-a8ff-000c29c89e42 | mysql_3     |        3306 | ONLINE       |
+---------------------------+--------------------------------------+-------------+-------------+--------------+
3 rows in set (0.00 sec)

此刻去2个secondary节点读取test.a_1表,表是不存在的。

secondary 1:
[root@mysql.sock][test]>select * from test.a_1;
ERROR 1146 (42S02): Table 'test.a_1' doesn't exist
[root@mysql.sock][test]>
secondary 2:
[root@mysql.sock][test]>select * from test.a_1;
ERROR 1146 (42S02): Table 'test.a_1' doesn't exist

error log输出信息:(set global log_error_verbosity = 3;)

2019-10-17T21:16:11.564211+08:00 10 [Warning] Slave SQL for channel
'group_replication_applier': Worker 1 failed executing transaction 'aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa:8' at master log ,
end_log_pos 220; Error executing row event: 'Table 'test.a_1' doesn't exist', Error_code: 1146

四、slave_skip_errors源码收效点

这个设置在Rows_log_event::do_apply_event 函数中收效,也便是DML Event开始运用的时分收效,这是惯例的SQL线程(或许Worker线程)调用的。

#ifdef HAVE_REPLICATION
  if (opt_slave_skip_errors)
    add_slave_skip_errors(opt_slave_skip_errors);
#endif
if (open_and_lock_tables(thd, rli->tables_to_lock, 0))//翻开表
    {
      uint actual_error= thd->get_stmt_da()->mysql_errno();
      if (thd->is_slave_error || thd->is_fatal_error)  
      {
        if (ignored_error_code(actual_error)) //这儿遭到 slave_skip_errors 参数操控 ignored_error_code会将slave_skip_errors的参数设置读取出来
        {
          if (log_warnings > 1)
            rli->report(WARNING_LEVEL, actual_error,
                        "Error executing row event: '%s'",
                        (actual_error ? thd->get_stmt_da()->message_text() :
                         "unexpected success or fatal error"));
          thd->get_stmt_da()->reset_condition_info(thd);
          clear_all_errors(thd, const_cast<Relay_log_info*>(rli));
          error= 0;
          goto end;
        }
        else
        {
          rli->report(ERROR_LEVEL, actual_error,
                      "Error executing row event: &apos;%s&apos;",
                      (actual_error ? thd->get_stmt_da()->message_text() :
                       "unexpected success or fatal error"));
          thd->is_slave_error= 1;
          const_cast<Relay_log_info*>(rli)->slave_close_thread_tables(thd);
          DBUG_RETURN(actual_error);
        }
      }
 ```
可以正常的看到MGR的实施逻辑遭到了该参数的影响。