本文内容遵从CC版权协议, 可以随意转载, 但必须以超链接形式标明文章原始出处和作者信息及版权声明网址: http://www.penglixun.com/tech/database/mysql_multi_slave_same_serverid.html 今天分析一个诡异问题，一个模拟Slave线程的程序，不断的被Master Ser
本文内容遵从CC版权协议, 可以随意转载, 但必须以超链接形式标明文章原始出处和作者信息及版权声明网址: http://www.penglixun.com/tech/database/mysql_multi_slave_same_serverid.html
今天分析一个诡异问题，一个模拟Slave线程的程序，不断的被Master Server给kill掉，最终发现是因为有两个Slave使用同样一个server id去连接Master Server，为什么两个Slave用同一个server id会被Master Server给Kill呢？分析了源码，这源于MySQL Replication的重连机制。
我们首先看看一个Slave注册到Master会发生什么，首先Slave需要向Master发送一个COM_REGISTER_SLAVE类型的请求（sql_parse.cc）命令请求，这里Master会使用register_slave函数注册一个Slave到slave_list。
<span style="color: #0000ff;">case</span> COM_REGISTER_SLAVE<span style="color: #008080;">:</span>
<span style="color: #008000;">{</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span><span style="color: #000040;">!</span>register_slave<span style="color: #008000;">(</span>thd, <span style="color: #008000;">(</span>uchar<span style="color: #000040;">*</span><span style="color: #008000;">)</span>packet, packet_length<span style="color: #008000;">)</span><span style="color: #008000;">)</span>
my_ok<span style="color: #008000;">(</span>thd<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
<span style="color: #0000ff;">break</span><span style="color: #008080;">;</span>
<span style="color: #008000;">}</span>在注册Slave线程的时候会发生什么呢？我们略去无用的代码直接看重点：（repl_failsafe.cc）
<span style="color: #0000ff;">int</span> register_slave<span style="color: #008000;">(</span>THD<span style="color: #000040;">*</span> thd, uchar<span style="color: #000040;">*</span> packet, uint packet_length<span style="color: #008000;">)</span>
<span style="color: #008000;">{</span>
<span style="color: #0000ff;">int</span> res<span style="color: #008080;">;</span>
SLAVE_INFO <span style="color: #000040;">*</span>si<span style="color: #008080;">;</span>
uchar <span style="color: #000040;">*</span>p<span style="color: #000080;">=</span> packet, <span style="color: #000040;">*</span>p_end<span style="color: #000080;">=</span> packet <span style="color: #000040;">+</span> packet_length<span style="color: #008080;">;</span>
.... <span style="color: #666666;">//省略</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span><span style="color: #000040;">!</span><span style="color: #008000;">(</span>si<span style="color: #000040;">-</span><span style="color: #000080;">></span>master_id<span style="color: #000080;">=</span> uint4korr<span style="color: #008000;">(</span>p<span style="color: #008000;">)</span><span style="color: #008000;">)</span><span style="color: #008000;">)</span>
si<span style="color: #000040;">-</span><span style="color: #000080;">></span>master_id<span style="color: #000080;">=</span> server_id<span style="color: #008080;">;</span>
si<span style="color: #000040;">-</span><span style="color: #000080;">></span>thd<span style="color: #000080;">=</span> thd<span style="color: #008080;">;</span>
pthread_mutex_lock<span style="color: #008000;">(</span><span style="color: #000040;">&</span>LOCK_slave_list<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
unregister_slave<span style="color: #008000;">(</span>thd,<span style="color: #0000dd;">0</span>,<span style="color: #0000dd;">0</span><span style="color: #008000;">)</span><span style="color: #008080;">;</span> <span style="color: #666666;">//关键在这里，先取消注册server_id相同的Slave线程</span>
res<span style="color: #000080;">=</span> my_hash_insert<span style="color: #008000;">(</span><span style="color: #000040;">&</span>slave_list, <span style="color: #008000;">(</span>uchar<span style="color: #000040;">*</span><span style="color: #008000;">)</span> si<span style="color: #008000;">)</span><span style="color: #008080;">;</span> <span style="color: #666666;">//把新的Slave线程注册到slave_list</span>
pthread_mutex_unlock<span style="color: #008000;">(</span><span style="color: #000040;">&</span>LOCK_slave_list<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
<span style="color: #0000ff;">return</span> res<span style="color: #008080;">;</span>
.....
<span style="color: #008000;">}</span>这是什么意思呢？这就是重连机制，slave_list是一个Hash表，server_id是Key，每一个线程注册上来，需要删掉同样server_id的Slave线程，再把新的Slave线程加到slave_list表中。
线程注册上来后，请求Binlog，发送COM_BINLOG_DUMP请求，Master会发送binlog给Slave，代码如下：
<span style="color: #0000ff;">case</span> COM_BINLOG_DUMP<span style="color: #008080;">:</span>
<span style="color: #008000;">{</span>
ulong pos<span style="color: #008080;">;</span>
ushort flags<span style="color: #008080;">;</span>
uint32 slave_server_id<span style="color: #008080;">;</span>
status_var_increment<span style="color: #008000;">(</span>thd<span style="color: #000040;">-</span><span style="color: #000080;">></span>status_var.<span style="color: #007788;">com_other</span><span style="color: #008000;">)</span><span style="color: #008080;">;</span>
thd<span style="color: #000040;">-</span><span style="color: #000080;">></span>enable_slow_log<span style="color: #000080;">=</span> opt_log_slow_admin_statements<span style="color: #008080;">;</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span>check_global_access<span style="color: #008000;">(</span>thd, REPL_SLAVE_ACL<span style="color: #008000;">)</span><span style="color: #008000;">)</span>
<span style="color: #0000ff;">break</span><span style="color: #008080;">;</span>
<span style="color: #ff0000; font-style: italic;">/* TODO: The following has to be changed to an 8 byte integer */</span>
pos <span style="color: #000080;">=</span> uint4korr<span style="color: #008000;">(</span>packet<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
flags <span style="color: #000080;">=</span> uint2korr<span style="color: #008000;">(</span>packet <span style="color: #000040;">+</span> <span style="color: #0000dd;">4</span><span style="color: #008000;">)</span><span style="color: #008080;">;</span>
thd<span style="color: #000040;">-</span><span style="color: #000080;">></span>server_id<span style="color: #000080;">=</span><span style="color: #0000dd;">0</span><span style="color: #008080;">;</span> <span style="color: #ff0000; font-style: italic;">/* avoid suicide */</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span><span style="color: #008000;">(</span>slave_server_id<span style="color: #000080;">=</span> uint4korr<span style="color: #008000;">(</span>packet<span style="color: #000040;">+</span><span style="color: #0000dd;">6</span><span style="color: #008000;">)</span><span style="color: #008000;">)</span><span style="color: #008000;">)</span> <span style="color: #666666;">// mysqlbinlog.server_id==0</span>
kill_zombie_dump_threads<span style="color: #008000;">(</span>slave_server_id<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
thd<span style="color: #000040;">-</span><span style="color: #000080;">></span>server_id <span style="color: #000080;">=</span> slave_server_id<span style="color: #008080;">;</span>
general_log_print<span style="color: #008000;">(</span>thd, command, <span style="color: #FF0000;">"Log: '%s' Pos: %ld"</span>, packet<span style="color: #000040;">+</span><span style="color: #0000dd;">10</span>,
<span style="color: #008000;">(</span><span style="color: #0000ff;">long</span><span style="color: #008000;">)</span> pos<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
mysql_binlog_send<span style="color: #008000;">(</span>thd, thd<span style="color: #000040;">-</span><span style="color: #000080;">></span>strdup<span style="color: #008000;">(</span>packet <span style="color: #000040;">+</span> <span style="color: #0000dd;">10</span><span style="color: #008000;">)</span>, <span style="color: #008000;">(</span>my_off_t<span style="color: #008000;">)</span> pos, flags<span style="color: #008000;">)</span><span style="color: #008080;">;</span> <span style="color: #666666;">//不断的发送日志给slave端</span>
unregister_slave<span style="color: #008000;">(</span>thd,<span style="color: #0000dd;">1</span>,<span style="color: #0000dd;">1</span><span style="color: #008000;">)</span><span style="color: #008080;">;</span> <span style="color: #666666;">//发送完成后清理Slave线程，因为执行到这一步肯定是binlog dump线程被kill了</span>
<span style="color: #ff0000; font-style: italic;">/* fake COM_QUIT -- if we get here, the thread needs to terminate */</span>
error <span style="color: #000080;">=</span> TRUE<span style="color: #008080;">;</span>
<span style="color: #0000ff;">break</span><span style="color: #008080;">;</span>
<span style="color: #008000;">}</span>mysql_binlog_send函数在sql_repl.cc，里面是轮询Master binlog，发送给Slave。
再来简单看看unregister_slave做了什么（repl_failsafe.cc）：
<span style="color: #0000ff;">void</span> unregister_slave<span style="color: #008000;">(</span>THD<span style="color: #000040;">*</span> thd, <span style="color: #0000ff;">bool</span> only_mine, <span style="color: #0000ff;">bool</span> need_mutex<span style="color: #008000;">)</span>
<span style="color: #008000;">{</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span>thd<span style="color: #000040;">-</span><span style="color: #000080;">></span>server_id<span style="color: #008000;">)</span>
<span style="color: #008000;">{</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span>need_mutex<span style="color: #008000;">)</span>
pthread_mutex_lock<span style="color: #008000;">(</span><span style="color: #000040;">&</span>LOCK_slave_list<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
SLAVE_INFO<span style="color: #000040;">*</span> old_si<span style="color: #008080;">;</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span><span style="color: #008000;">(</span>old_si <span style="color: #000080;">=</span> <span style="color: #008000;">(</span>SLAVE_INFO<span style="color: #000040;">*</span><span style="color: #008000;">)</span>hash_search<span style="color: #008000;">(</span><span style="color: #000040;">&</span>slave_list,
<span style="color: #008000;">(</span>uchar<span style="color: #000040;">*</span><span style="color: #008000;">)</span><span style="color: #000040;">&</span>thd<span style="color: #000040;">-</span><span style="color: #000080;">></span>server_id, <span style="color: #0000dd;">4</span><span style="color: #008000;">)</span><span style="color: #008000;">)</span> <span style="color: #000040;">&&</span>
<span style="color: #008000;">(</span><span style="color: #000040;">!</span>only_mine <span style="color: #000040;">||</span> old_si<span style="color: #000040;">-</span><span style="color: #000080;">></span>thd <span style="color: #000080;">==</span> thd<span style="color: #008000;">)</span><span style="color: #008000;">)</span> <span style="color: #666666;">//拿到slave值</span>
hash_delete<span style="color: #008000;">(</span><span style="color: #000040;">&</span>slave_list, <span style="color: #008000;">(</span>uchar<span style="color: #000040;">*</span><span style="color: #008000;">)</span>old_si<span style="color: #008000;">)</span><span style="color: #008080;">;</span> <span style="color: #666666;">//从slave_list中拿掉</span>
<span style="color: #0000ff;">if</span> <span style="color: #008000;">(</span>need_mutex<span style="color: #008000;">)</span>
pthread_mutex_unlock<span style="color: #008000;">(</span><span style="color: #000040;">&</span>LOCK_slave_list<span style="color: #008000;">)</span><span style="color: #008080;">;</span>
<span style="color: #008000;">}</span>
<span style="color: #008000;">}</span>这就可以解释同样的server_id为什么会被kill，因为一旦注册上去，就会现删除相同server_id的Slave线程，然后把当前的Slave加入，这是因为有时Slave断开了，重新请求上来，当然需要踢掉原来的线程，这就是线程重连机制。
切记，一个MySQL集群中，绝不可以出现相同server_id的实例，否则各种诡异的问题可是接踵而来。
MySQL ändert den Namen der Datentabelle
MySQL erstellt eine gespeicherte Prozedur
Der Unterschied zwischen Mongodb und MySQL
So überprüfen Sie, ob das MySQL-Passwort vergessen wurde
MySQL-Datenbank erstellen
MySQL-Standard-Transaktionsisolationsstufe
Der Unterschied zwischen SQL Server und MySQL
mysqlPasswort vergessen
