MySQL手动注册binlog文件造成主从异常的原因

一、问题来源

有一个朋友@水米田问我，基于POSITION的主从。他做了如下的操作

将备份的一些binlog文件加入到了目录中
修改index文件，加入了这些binlog文件
flush binary logs
然后整个主从环境大量延迟。

二、朋友的测试

下面是另外一个朋友@徐晨亮的测试：

master上：
(root:db1@xucl:10:30:22)[(none)]> show binary logs;
+---------------------+-----------+
| Log_name      | File_size |
+---------------------+-----------+
| mysql-binlog.000031 |   1019 |
| mysql-binlog.000032 |    424 |
| mysql-binlog.000033 |    244 |
| mysql-binlog.000034 |   2332 |
| mysql-binlog.000035 |   2134 |
| mysql-binlog.000036 |  845915 |
| mysql-binlog.000037 |   11735 |
| mysql-binlog.000038 |    284 |
| mysql-binlog.000039 |    284 |
| mysql-binlog.000040 |    284 |
| mysql-binlog.000041 |    284 |
| mysql-binlog.000042 |    234 |
+---------------------+-----------+
12 rows in set (0.00 sec)
(root:db1@xucl:10:30:34)[(none)]> purge binary logs to 'mysql-binlog.000039';
Query OK, 0 rows affected (0.00 sec)
(root:db1@xucl:10:30:49)[(none)]> show binary logs;
+---------------------+-----------+
| Log_name      | File_size |
+---------------------+-----------+
| mysql-binlog.000039 |    284 |
| mysql-binlog.000040 |    284 |
| mysql-binlog.000041 |    284 |
| mysql-binlog.000042 |    234 |
+---------------------+-----------+
4 rows in set (0.00 sec)
执行插入数据
(root:db1@xucl:10:31:23)[xucl]> insert into t values(9,9);
将备份的binlog拷贝回原先的目录并修改index文件进行注册
[root@izbp12nspj47ypto9t6vyez logs]# ll
总用量 884
-rw-r----- 1 mysql mysql  1019 5月 20 22:03 mysql-binlog.000031
-rw-r----- 1 mysql mysql  424 5月 20 22:03 mysql-binlog.000032
-rw-r----- 1 mysql mysql  244 5月 20 22:03 mysql-binlog.000033
-rw-r----- 1 mysql mysql  2332 5月 20 22:03 mysql-binlog.000034
-rw-r----- 1 mysql mysql  2134 5月 20 22:03 mysql-binlog.000035
-rw-r----- 1 mysql mysql 845915 5月 20 22:03 mysql-binlog.000036
-rw-r----- 1 mysql mysql 11735 5月 20 22:05 mysql-binlog.000037
-rw-r----- 1 mysql mysql  284 5月 20 22:06 mysql-binlog.000038
-rw-r----- 1 mysql mysql  284 5月 21 10:28 mysql-binlog.000039
-rw-r----- 1 mysql mysql  284 5月 21 10:28 mysql-binlog.000040
-rw-r----- 1 mysql mysql  284 5月 21 10:28 mysql-binlog.000041
-rw-r----- 1 mysql mysql  491 5月 21 10:31 mysql-binlog.000042
-rw-r----- 1 mysql mysql  204 5月 21 10:30 mysql-binlog.index
主库flush binary logs
(root:db1@xucl:10:32:51)[(none)]> flush binary logs;
Query OK, 0 rows affected (0.01 sec)
(root:db1@xucl:10:32:57)[(none)]> show binary logs;
+---------------------+-----------+
| Log_name      | File_size |
+---------------------+-----------+
| mysql-binlog.000031 |   1019 |
| mysql-binlog.000032 |    424 |
| mysql-binlog.000033 |    244 |
| mysql-binlog.000034 |   2332 |
| mysql-binlog.000035 |   2134 |
| mysql-binlog.000036 |  845915 |
| mysql-binlog.000037 |   11735 |
| mysql-binlog.000038 |    284 |
| mysql-binlog.000039 |    284 |
| mysql-binlog.000040 |    284 |
| mysql-binlog.000041 |    284 |
| mysql-binlog.000042 |    541 |
| mysql-binlog.000043 |    234 |
+---------------------+-----------+
13 rows in set (0.00 sec)
此时，slave报错如下：
(root:db1@xucl:10:31:05)[(none)]> show slave status\G
*************************** 1. row ***************************
        Slave_IO_State:
         Master_Host: 127.0.0.1
         Master_User: repl
         Master_Port: 3306
        Connect_Retry: 60
       Master_Log_File: mysql-binlog.000035
     Read_Master_Log_Pos: 194
        Relay_Log_File: izbp12nspj47ypto9t6vyez-relay-bin.000011
        Relay_Log_Pos: 373
    Relay_Master_Log_File: mysql-binlog.000035
       Slave_IO_Running: No
      Slave_SQL_Running: Yes
       Replicate_Do_DB:
     Replicate_Ignore_DB:
      Replicate_Do_Table:
    Replicate_Ignore_Table:
   Replicate_Wild_Do_Table:
 Replicate_Wild_Ignore_Table:
          Last_Errno: 0
          Last_Error:
         Skip_Counter: 0
     Exec_Master_Log_Pos: 194
       Relay_Log_Space: 648
       Until_Condition: None
        Until_Log_File:
        Until_Log_Pos: 0
      Master_SSL_Allowed: No
      Master_SSL_CA_File:
      Master_SSL_CA_Path:
       Master_SSL_Cert:
      Master_SSL_Cipher:
        Master_SSL_Key:
    Seconds_Behind_Master: NULL
Master_SSL_Verify_Server_Cert: No
        Last_IO_Errno: 1236
        Last_IO_Error: Got fatal error 1236 from master when reading data from binary log: 'Cannot replicate GTID-transaction when @@GLOBAL.GTID_MODE = OFF, at file /storage/single/mysql3306/logs/mysql-binlog.000035, position 194.; the first event 'mysql-binlog.000039' at 234, the last event read from '/storage/single/mysql3306/logs/mysql-binlog.000035' at 259, the last byte read from '/storage/single/mysql33'
        Last_SQL_Errno: 0
        Last_SQL_Error:
 Replicate_Ignore_Server_Ids:
       Master_Server_Id: 3306
         Master_UUID: e8bdf01a-c79b-11e8-82b3-00163e088352
       Master_Info_File: mysql.slave_master_info
          SQL_Delay: 0
     SQL_Remaining_Delay: NULL
   Slave_SQL_Running_State: Slave has read all relay log; waiting for more updates
      Master_Retry_Count: 86400
         Master_Bind:
   Last_IO_Error_Timestamp: 190521 10:32:57
   Last_SQL_Error_Timestamp:
        Master_SSL_Crl:
      Master_SSL_Crlpath:
      Retrieved_Gtid_Set:
      Executed_Gtid_Set: 4c423515-6661-11e9-b767-00163e088352:1-7,
e8bdf01a-c79b-11e8-82b3-00163e088352:1-57192
        Auto_Position: 0
     Replicate_Rewrite_DB:
         Channel_Name:
      Master_TLS_Version:
1 row in set (0.00 sec)
从slave上的报错来看，在主库flush binary logs后，从库又读取注册的binlog并且又apply了

三、现象说明

从整个测试来看，MySQL视乎将手动注册的文件进行了传输和应用。报错是因为这个库以前是GITD_MODE=ON的，但是测试的时候已经关闭了GTID_MODE，改为了POSITION的模式，这个报错是因为DUMP线程会进行检测：

这个图来自我新写的一个系列（暂时还没发布，大概年底才能写好）。不管怎么说，DUMP线程已经在开始传输老的binlog文件了。那么原因是什么呢？下面我们将进行解释。

四、flush binary logs对binlog的操作

flush binary logs 将包含如下操作：

获取新的binlog文件名字
关闭旧的binlog
关闭index file
打开index file
打开新的binlog
将新的binlog加入到indexfile

具体可以参考函数MYSQL_BIN_LOG::new_file_impl。其中要说明一下获取新的binlog文件名字是通过函数find_uniq_filename实现的，其中包含如下代码：

 file_info= dir_info->dir_entry;
 for (i= dir_info->number_off_files ; i-- ; file_info++)
 {
  if (strncmp(file_info->name, start, length) == 0 &&
  is_number(file_info->name+length, &number,0))
  {
   set_if_bigger(max_found, number);
  }
 }
...
 *next= (need_next || max_found == 0) ? max_found + 1 : max_found;

大概意思就是扫描index file文件里面的binlog文件，获取其序号最高的一个，然后加1。栈帧如下：

#0 find_uniq_filename (name=0x7fffec5ec6d0 "/mysqldata/mysql3340/log/binlog", next=0x7fffec5ec678, need_next=true)
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:3679
#1 0x000000000187d208 in generate_new_log_name (new_name=0x7fffec5ec6d0 "/mysqldata/mysql3340/log/binlog", new_ext=0x0, 
  log_name=0x7ffedc011520 "/mysqldata/mysql3340/log/binlog", is_binlog=true) at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:3767
#2 0x0000000001884fdb in MYSQL_BIN_LOG::new_file_impl (this=0x2e83640, need_lock_log=false, extra_description_event=0x0)
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:7175
#3 0x0000000001884cbb in MYSQL_BIN_LOG::new_file_without_locking (this=0x2e83640, extra_description_event=0x0)
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:7099
#4 0x0000000001886b6b in MYSQL_BIN_LOG::rotate (this=0x2e83640, force_rotate=true, check_purge=0x7fffec5ecbfb)
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:7775
#5 0x0000000001886d53 in MYSQL_BIN_LOG::rotate_and_purge (this=0x2e83640, thd=0x7ffedc000b90, force_rotate=true)
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:7846

因此即便我们手动修改了index file，flush binary logs却不会有问题，因为它实际扫描了index file文件。
同时我们也看到flush binary logs重新加载了index file，这个时候手动修改的index file就生效了，使用show binary logs就能查看到你加入的文件了。

五、主从问题的产生

binlog切换后，DUMP线程也需要读取下一个binlog文件。我们来看看它是怎么确认读取哪一个文件的。

在函数sender.run()中可以找到循环每个binlog文件的代码。下面一句是寻找下一个binlog文件：

int error= mysql_bin_log.find_next_log(&m_linfo, 0);

mysql_bin_log.find_next_log 包含的代码：

 my_b_seek(&index_file, linfo->index_file_offset);//进行偏移量偏移

 linfo->index_file_start_offset= linfo->index_file_offset;
 length=my_b_gets(&index_file, fname, FN_REFLEN));//读取文件名字
...
  if(normalize_binlog_name(full_fname, fname, is_relay_log))
...

 linfo->index_file_offset= my_b_tell(&index_file);//偏移量从新记录以备下一次使用

我们能够看到DUMP线程并没有实际扫描整个index文件，而是通过一个index文件的偏移量进行读取。如果手动修改index文件那么偏移量就出现了错乱。因此DUMP发送的下一个文件将是不确定的。因此出现了发送手动注册的binlog文件给从库的现象，这种情况下可能出现下面情况：

如果是GTID_MODE 关闭，使用POSITION那么这种情况一定报错，比如重复的行。
如果是GTID_MODE 和AUTO_POSITION=1，那么DUMP线程会进行GTID的过滤不发送，因为Event不发送所以延迟会持续一段时间为0。
如果是GTID_MODE 和AUTO_POSITION=0，那么SQL线程应用GITD_EVENT的时候会进行过滤，延迟可能出现很大的情况。

尽管GTID可能可以屏蔽这种问题，但是DUMP线程已经在考虑发送老的binlog文件了，这是不合适的。

六、purge binary logs能够维护这个偏移量

为什么purge binary logs不会出现问题呢，因为在purge binary logs的语句下，会维护这个偏移量如下：

 virtual void operator()(THD *thd)
 {
  LOG_INFO* linfo;
  mysql_mutex_lock(&thd->LOCK_thd_data);
  if ((linfo= thd->current_linfo))//b binlog.cc:2829
  {
   /*
    Index file offset can be less that purge offset only if
    we just started reading the index file. In that case
    we have nothing to adjust.
   */
   if (linfo->index_file_offset < m_purge_offset)
    linfo->fatal = (linfo->index_file_offset != 0);
   else
    linfo->index_file_offset -= m_purge_offset;
  }
  mysql_mutex_unlock(&thd->LOCK_thd_data);

我们可以看到linfo->index_file_offset -= m_purge_offset;这样一个语句。下面是栈帧：

#0 Adjust_offset::operator() (this=0x7fffec5ec720, thd=0x7ffedc000be0) at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:2831
#1 0x0000000000eef320 in Do_THD::operator() (this=0x7fffec5ec6a0, thd=0x7ffedc000be0) at /mysqldata/percona-server-locks-detail-5.7.22/sql/mysqld_thd_manager.cc:46
#2 0x0000000000eefa0f in std::for_each (__first=0x3003358, __last=0x3003368, __f=...)
  at /usr/lib/gcc/x86_64-redhat-linux/4.4.7/../../../../include/c++/4.4.7/bits/stl_algo.h:4200
#3 0x0000000000eeefc0 in Global_THD_manager::do_for_all_thd (this=0x3003340, func=0x7fffec5ec720)
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/mysqld_thd_manager.cc:273
#4 0x000000000187ae0a in adjust_linfo_offsets (purge_offset=0) at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:2873
#5 0x0000000001883239 in MYSQL_BIN_LOG::remove_logs_from_index (this=0x2e83640, log_info=0x7fffec5ec7d0, need_update_threads=true)
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:6352
#6 0x0000000001883676 in MYSQL_BIN_LOG::purge_logs (this=0x2e83640, to_log=0x7fffec5eca80 "/mysqldata/mysql3340/log/binlog.000001", included=false, 
  need_lock_index=true, need_update_threads=true, decrease_log_space=0x0, auto_purge=false) at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:6469
#7 0x000000000187b4b5 in purge_master_logs (thd=0x7ffee0000c00, to_log=0x7ffee0006600 "binlog.000001")
  at /mysqldata/percona-server-locks-detail-5.7.22/sql/binlog.cc:3127

七、测试POSITION模式下的报错

1、环境

mysql> show binary logs;
+---------------+-----------+
| Log_name   | File_size |
+---------------+-----------+
| binlog.000001 |    198 |
| binlog.000002 |    154 |
+---------------+-----------+
2 rows in set (0.00 sec)

mysql> show create table testcp \G;
*************************** 1. row ***************************
    Table: testcp
Create Table: CREATE TABLE `testcp` (
 `id` int(11) NOT NULL,
 PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8
1 row in set (0.00 sec)

ERROR: 
No query specified

2、执行一个语句

主库：

mysql> insert into testcp values(20);
Query OK, 1 row affected (0.43 sec)

mysql> select * from testcp;
+----+
| id |
+----+
| 10 |
| 20 |
+----+
2 rows in set (0.01 sec)

从库：

mysql> show slave status \G;
*************************** 1. row ***************************
        Slave_IO_State: Waiting for master to send event
         Master_Host: 192.168.99.41
         Master_User: repl
         Master_Port: 3340
        Connect_Retry: 60
       Master_Log_File: binlog.000002
     Read_Master_Log_Pos: 417
        Relay_Log_File: relay.000004
        Relay_Log_Pos: 624
    Relay_Master_Log_File: binlog.000002
       Slave_IO_Running: Yes
      Slave_SQL_Running: Yes
...
mysql> select * from testcp;
+----+
| id |
+----+
| 10 |
| 20 |
+----+

这个时候DUMP线程index file偏移指针如下：

3、主库执行purge binary logs

做之前拷贝原有binlog.000001为binlog.000001bak因为等会要拷贝回去

mysql> purge binary logs to 'binlog.000002';
Query OK, 0 rows affected (3.14 sec)

mysql> show binary logs;
+---------------+-----------+
| Log_name   | File_size |
+---------------+-----------+
| binlog.000002 |    417 |
+---------------+-----------+
1 row in set (0.00 sec)

因为purge binary logs命令会维护偏移指针，这个时候DUMP线程的index file偏移指针如下：

4、手动更改

将binlog.000001bak拷贝为binlog.000001，然后修改index file将binlog.000001加入回去，然后flush binary logs如下：

mysql> flush binary logs;
Query OK, 0 rows affected (0.15 sec)

mysql> show binary logs;
+---------------+-----------+
| Log_name   | File_size |
+---------------+-----------+
| binlog.000001 |     198 |
| binlog.000002 |    461 |
+---------------+-----------+
2 rows in set (0.00 sec)

手动完成这个工作并不会维护DUMP线程的index file偏移指针，因此如下：

就这样DUMP线程重新发送了一次binlog.000002的内容，POSITION的从库只能报错了如下：

mysql> select * from replication_applier_status_by_worker \G
*************************** 1. row ***************************
     CHANNEL_NAME: 
      WORKER_ID: 1
      THREAD_ID: NULL
    SERVICE_STATE: OFF
LAST_SEEN_TRANSACTION: ANONYMOUS
  LAST_ERROR_NUMBER: 1062
  LAST_ERROR_MESSAGE: Worker 1 failed executing transaction 'ANONYMOUS' at master log binlog.000002, end_log_pos 386; Could not execute Write_rows event on table testmts.testcp; Duplicate entry '20' for key 'PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log binlog.000002, end_log_pos 386
 LAST_ERROR_TIMESTAMP: 2019-05-21 14:46:26

八、总结

这种操作非常不规范，如果实在要这么做考虑如下步骤：