当先锋百科网

首页 1 2 3 4 5 6 7

一、两个不同的现象

首先建立一张有几条数据的表就可以了,我这里是baguait1表了。

SESSION1SESSION2SESSION3
步骤1:select sleep(1000) from baguait1 for update;  
 步骤2:flush table with read lock;堵塞 
  步骤3:kill session2
  

步骤4:select * from baguait1 limit 1;成

步骤2 “flush table with read lock;”操作等待状态为“Waiting for global read lock”,如下:

mysql> select Id,State,Info   from information_schema.processlist  where command<>'sleep';
+----+------------------------------+------------------------------------------------------------------------------------+
| Id | State                        | Info                                                                               |
+----+------------------------------+------------------------------------------------------------------------------------+
|  1 | Waiting on empty queue       | NULL                                                                               |
| 18 | Waiting for global read lock | flush table with read lock                                                         |
|  3 | User sleep                   | select sleep(1000) from baguait1 for update                                        |
|  6 | executing                    | select Id,State,Info   from information_schema.processlist  where command<>'sleep' |
+----+------------------------------+------------------------------------------------------------------------------------+
  • 案例2

这里比较奇怪了,实际上我很久以前就遇到过和测试过但是没有仔细研究过,这次刚好详细看看。

SESSION1SESSION2SESSION3
步骤1:select sleep(1000) from baguait1  
 步骤2:flush table with read lock;堵塞 
  步骤3:kill session2
  步骤4:select * from baguait1 limit 1;堵塞

步骤2 “flush table with read lock;”操作等待状态为 “Waiting for table flush”,状态如下:

mysql> select Id,State,Info   from information_schema.processlist  where command<>'sleep';
+----+-------------------------+------------------------------------------------------------------------------------+
| Id | State                   | Info                                                                               |
+----+-------------------------+------------------------------------------------------------------------------------+
|  1 | Waiting on empty queue  | NULL                                                                               |
| 26 | User sleep              | select sleep(1000) from baguait1                                                   |
| 23 | Waiting for table flush | flush table with read lock                                                         |
|  6 | executing               | select Id,State,Info   from information_schema.processlist  where command<>'sleep' |
+----+-------------------------+------------------------------------------------------------------------------------+

步骤4 “select * from testmts.baguait1 limit 1”操作等待状态为 “Waiting for table flush”,这个想象看起来非常奇怪没有任何特殊的其他操作,select居然堵塞了。

mysql> select Id,State,Info   from information_schema.processlist  where command<>'sleep';
+----+-------------------------+------------------------------------------------------------------------------------+
| Id | State                   | Info                                                                               |
+----+-------------------------+------------------------------------------------------------------------------------+
|  1 | Waiting on empty queue  | NULL                                                                               |
| 26 | User sleep              | select sleep(1000) from baguait1                                                   |
| 27 | executing               | select Id,State,Info   from information_schema.processlist  where command<>'sleep' |
|  6 | Waiting for table flush | select * from testmts.baguait1 limit 1                                             |
+----+-------------------------+------------------------------------------------------------------------------------+

如果仔细对比两个案例实际上区别仅仅在于 步骤1中的select 语句是否加了for update,案例2中我们发现即便我们将“flush table with read lock;”会话KILL掉也会堵塞随后的关于本表上全部操作(包括select),这个等待实际上会持续到步骤1的sleep操作完成过后。

对于线上数据库的话,如果在长时间的select大表期间执行“flush table with read lock;”就会出现这种情况,这种情况会造成全部关于本表的操作等待,即便你发现后杀掉了FTWRL会话也无济于事,等待会持续到select操作完成后,除非你KILL掉长时间的select操作。

为什么会出现这种情况呢?我们接下来慢慢分析。

二、sleep 函数生效点

关于本案例中我使用sleep函数来代替select 大表操作做为测试,在这里这个代替是成立的。为什么成立呢我们来看一下sleep函数的生效点如下:

T@3: | | | | | | | | >evaluate_join_record
T@3: | | | | | | | | | enter: join: 0x7ffee0007350 join_tab index: 0 table: tii cond: 0x0
T@3: | | | | | | | | | counts: evaluate_join_record join->examined_rows++: 1
T@3: | | | | | | | | | >end_send
T@3: | | | | | | | | | | >Query_result_send::send_data
T@3: | | | | | | | | | | | >send_result_set_row
T@3: | | | | | | | | | | | | >THD::enter_cond
T@3: | | | | | | | | | | | | | THD::enter_stage: 'User sleep' /mysqldata/percona-server-locks-detail-5.7.22/sql/item_func.cc:6057
T@3: | | | | | | | | | | | | | >PROFILING::status_change
T@3: | | | | | | | | | | | | | <PROFILING::status_change 384
T@3: | | | | | | | | | | | | <THD::enter_cond 3405

这里看出sleep的生效点实际上每次Innodb层返回一行数据经过where条件判断后,再触发sleep函数,也就是每行经过where条件过滤的数据在发送给客户端之前都会进行一次sleep操作。这个时候实际上该打开表的和该上MDL LOCK的都已经完成了,因此使用sleep函数来模拟大表select操作导致的FTWRL堵塞是可以的。

三、FTWRL做了什么工作

实际上这部分我们可以在函数mysql_execute_command寻找case SQLCOM_FLUSH 的部分,实际上主要调用函数为reload_acl_and_cache,其中核心部分为:

if (thd->global_read_lock.lock_global_read_lock(thd))//加 MDL GLOBAL 级别S锁
    return 1;                               // Killed
      if (close_cached_tables(thd, tables, //关闭表操作释放 share 和 cache
                              ((options & REFRESH_FAST) ?  FALSE : TRUE),
                              thd->variables.lock_wait_timeout)) //等待时间受lock_wait_timeout影响
      {
        /*
          NOTE: my_error() has been already called by reopen_tables() within
          close_cached_tables().
        */
        result= 1;
      }
      if (thd->global_read_lock.make_global_read_lock_block_commit(thd)) // MDL COMMIT 锁
      {
        /* Don't leave things in a half-locked state */
        thd->global_read_lock.unlock_global_read_lock(thd);
        return 1;
      }

更具体的关闭表的操作和释放table缓存的部分包含在函数close_cached_tables中,我就不详细写了。但是我们需要明白table缓存实际上包含两个部分:

  • table cache define:每一个表第一次打开的时候都会建立一个静态的表定义结构内存,当多个会话同时访问同一个表的时候,从这里拷贝成相应的instance供会话自己使用。由参数table_definition_cache定义大小,由状态值Open_table_definitions查看当前使用的个数。对应函数get_table_share。
  • table cache instance:同上所述,这是会话实际使用的表定义结构是一个instance。由参数table_open_cache定义大小,由状态值Open_tables查看当前使用的个数。对应函数open_table_from_share。

这里我统称为table缓存,好了下面是我总结的FTWRl的大概步骤:

第一步: 加MDL LOCK类型为GLOBAL 级别为S。如果出现等待状态为‘Waiting for global read lock’。注意select语句不会上GLOBAL级别上锁,但是DML/DDL/FOR UPDATE语句会上GLOBAL级别的IX锁,IX锁和S锁不兼容会出现这种等待。下面是这个兼容矩阵:

          | Type of active   |
  Request |   scoped lock    |
   type   | IS(*)  IX   S  X |
 ---------+------------------+
 IS       |  +      +   +  + |
 IX       |  +      +   -  - |
 S        |  +      -   +  - |
 X        |  +      -   -  - |

第二步:推进全局表缓存版本。源码中就是一个全局变量 refresh_version++。
第三步:释放没有使用的table 缓存。可自行参考函数close_cached_tables函数。
第四步:判断是否有正在占用的table缓存,如果有则等待,等待占用者释放。等待状态为’Waiting for table flush’。这一步会去判断table缓存的版本和全局表缓存版本是否匹配,如果不匹配则等待如下:

for (uint idx=0 ; idx < table_def_cache.records ; idx++) 
      {
        share= (TABLE_SHARE*) my_hash_element(&table_def_cache, idx); //寻找整个 table cache shared hash结构
        if (share->has_old_version()) //如果版本 和 当前 的 refresh_version 版本不一致
        {
          found= TRUE;
          break; //跳出第一层查找 是否有老版本 存在
        }
      }
...
if (found)//如果找到老版本,需要等待
    {
      /*
        The method below temporarily unlocks LOCK_open and frees
        share's memory.
      */
      if (share->wait_for_old_version(thd, &abstime,
                                    MDL_wait_for_subgraph::DEADLOCK_WEIGHT_DDL))
      {
        mysql_mutex_unlock(&LOCK_open);
        result= TRUE;
        goto err_with_reopen;
      }
    }

而等待的结束就是占用的table缓存的占用者释放,这个释放操作存在于函数close_thread_table中,如下:

if (table->s->has_old_version() || table->needs_reopen() ||
      table_def_shutdown_in_progress)
  {
    tc->remove_table(table);//关闭 table cache instance
    mysql_mutex_lock(&LOCK_open);
    intern_close_table(table);//去掉 table cache define
    mysql_mutex_unlock(&LOCK_open);
  }

最终会调用函数MDL_wait::set_status将FTWRL唤醒,也就是说对于正在占用的table缓存释放者不是FTWRL会话而是占用者自己。不管怎么样最终整个table缓存将会被清空,如果经过FTWRL后去查看Open_table_definitions和Open_tables将会发现重新计数了。下面是唤醒函数的代码,也很明显:

bool MDL_wait::set_status(enum_wait_status status_arg) open_table
{
  bool was_occupied= TRUE;
  mysql_mutex_lock(&m_LOCK_wait_status);
  if (m_wait_status == EMPTY)
  {
    was_occupied= FALSE;
    m_wait_status= status_arg;
    mysql_cond_signal(&m_COND_wait_status);//唤醒
  }
  mysql_mutex_unlock(&m_LOCK_wait_status);//解锁
  return was_occupied;
}

第五步:加MDL LOCK类型COMMIT 级别为S。如果出现等待状态为‘Waiting for commit lock’。如果有大事务的提交很可能出现这种等待。

四、案例1解析

步骤1 我们使用select for update语句,这个语句会加GLOBAL级别的IX锁,持续到语句结束(注意实际上还会加对象级别的MDL_SHARED_WRITE(SW)锁持续到事务结束,和FTWRL无关不做描述)

步骤2 我们使用FTWRL语句,根据上面的分析需要获取GLOBAL级别的S锁,不兼容,因此出现了等待‘Waiting for global read lock’

步骤3 我们KILL掉了FTWRL会话,这种情况下会话退出,FTWRL就像没有执行过一样不会有任何影响,因为它在第一步就堵塞了。

步骤4 我们的select操作不会受到任何影响

五、案例2解析

步骤1 我们使用select 语句,这个语句不会在GLOBAL级别上任何的锁(注意实际上还会加对象级别的MDL_SHARED_READ(SR)锁持续到事务结束,和FTWRL无关不做描述)

步骤2 我们使用FTWRL语句,根据上面的分析我们发现FTWRL语句可以获取了GLOBAL 级别的S锁,因为单纯的select 语句不会在GLOBAL级别上任何锁。同时会将全局表缓存版本推进然后释放掉没有使用的table 缓存,但是在第四步中会发现baguait1的表缓存正在被占用,因此出现了等待,等待状态为’Waiting for table flush’。

步骤3 我们KILL掉了FTWRL会话,这种情况下虽然GLOBAL 级别的S锁会释放,但是全局表缓存版本已经推进了,同时没有使用的table 缓存已经释放掉了。

步骤4 再次执行一个baguait1表上的select 查询操作,这个时候在打开表的时候会去判断是否table缓存的版本和全局表缓存版本匹配如果不匹配进入等待,等待为‘Waiting for table flush’,下面是这个判断:

if (share->has_old_version())
    {
      /*
        We already have an MDL lock. But we have encountered an old
        version of table in the table definition cache which is possible
        when someone changes the table version directly in the cache
        without acquiring a metadata lock (e.g. this can happen during
        "rolling" FLUSH TABLE(S)).
        Release our reference to share, wait until old version of
        share goes away and then try to get new version of table share.
      */
      release_table_share(share);
     ...
      wait_result= tdc_wait_for_old_version(thd, table_list->db,
                                            table_list->table_name,
                                            ot_ctx->get_timeout(),
                                            deadlock_weight);

整个等待操作和FTWRL一样,会等待占用者释放table缓存后才会醒来继续。

因此后续本表的所有select/DML/DDL都会堵塞,代价极高,即便KILL掉FTWRL会话也无用。

六、FTWRL堵塞和被堵塞的简单总结

(1)被什么堵塞

  • 长时间的DDL\DML\FOR UPDATE堵塞FTWRL,因为FTWRL需要获取 GLOBAL的S锁,而这些语句都会对GLOBAL持有IX(MDL_INTENTION_EXCLUSIVE)锁,根据兼容矩阵不兼容。等待为:Waiting for global read lock 。本文的案例1就是这种情况。
  • 长时间的select堵塞FTWRL, 因为FTWRL会释放所有空闲的table缓存,如果有占用者占用某些table缓存,则会等待占用者自己释放这些table缓存。等待为:Waiting for table flush 。本文的案例2就是这种情况,会堵塞随后关于本表的任何语句,即便KILL FTWRL会话也不行,除非KILL掉长时间的select操作才行。实际上flush table也会存在这种堵塞情况。
  • 长时间的commit(如大事务提交)也会堵塞FTWRL,因为FTWRL需要获取COMMIT的S锁,而commit语句会对commit持有IX(MDL_INTENTION_EXCLUSIVE)锁,根据兼容矩阵不兼容。

(2)堵塞什么

  • FTWRL会堵塞DDL\DML\FOR UPDATE操作,堵塞点为 GLOBAL级别 的S锁,等待为:Waiting for global read lock 。
  • FTWRL会堵塞commit操作,堵塞点为COMMIT的S锁,等待为Waiting for commit lock 。
  • FTWRL不会堵塞select操作,因为select不会在GLOBAL级别上锁。

最后提醒一下很多备份工具都要执行FTWRL操作,一定要注意它的堵塞场景和特殊场景。

 

备注栈帧和断点:

(1)使用的断点

  • MDL_context::acquire_lock 获取DML LOCK
  • open_table_from_share 获取table cache instance
  • alloc_table_share 分配table define(share)
  • get_table_share 获取table define(share)
  • close_cached_tables flush table关闭全部table cache instance 和table define
  • reload_acl_and_cache flush with read lock 进行MDL LOCK加锁为GLOBAL TYPE:S ,同时调用close_cached_tables 同时获取COMMIT级别 TYPE S
  • MDL_wait::set_status 唤醒操作
  • close_thread_table 占用者判断释放
  • my_hash_delete hash删除操作,从table cache instance 和table define中释放table缓存都是需要调用这个删除操作的。

(2)FTWRL堵塞栈帧由于select堵塞栈帧:

(gdb) bt
#0  0x00007ffff7bd3a5e in pthread_cond_timedwait@@GLIBC_2.3.2 () from /lib64/libpthread.so.0
#1  0x000000000192027b in native_cond_timedwait (cond=0x7ffedc007c78, mutex=0x7ffedc007c30, abstime=0x7fffec5bbb90)
    at /mysqldata/percona-server-locks-detail-5.7.22/include/thr_cond.h:129
#2  0x00000000019205ea in safe_cond_timedwait (cond=0x7ffedc007c78, mp=0x7ffedc007c08, abstime=0x7fffec5bbb90, 
    file=0x204cdd0 "/mysqldata/percona-server-locks-detail-5.7.22/sql/mdl.cc", line=1899) at /mysqldata/percona-server-locks-detail-5.7.22/mysys/thr_cond.c:88
#3  0x00000000014b9f21 in my_cond_timedwait (cond=0x7ffedc007c78, mp=0x7ffedc007c08, abstime=0x7fffec5bbb90, 
    file=0x204cdd0 "/mysqldata/percona-server-locks-detail-5.7.22/sql/mdl.cc", line=1899) at /mysqldata/percona-server-locks-detail-5.7.22/include/thr_cond.h:180
#4  0x00000000014ba484 in inline_mysql_cond_timedwait (that=0x7ffedc007c78, mutex=0x7ffedc007c08, abstime=0x7fffec5bbb90, 
    src_file=0x204cdd0 "/mysqldata/percona-server-locks-detail-5.7.22/sql/mdl.cc", src_line=1899)
    at /mysqldata/percona-server-locks-detail-5.7.22/include/mysql/psi/mysql_thread.h:1229
#5  0x00000000014bb702 in MDL_wait::timed_wait (this=0x7ffedc007c08, owner=0x7ffedc007b70, abs_timeout=0x7fffec5bbb90, set_status_on_timeout=true, 
    wait_state_name=0x2d897b0) at /mysqldata/percona-server-locks-detail-5.7.22/sql/mdl.cc:1899
#6  0x00000000016cdb30 in TABLE_SHARE::wait_for_old_version (this=0x7ffee0a4fc30, thd=0x7ffedc007b70, abstime=0x7fffec5bbb90, deadlock_weight=100)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/table.cc:4717
#7  0x000000000153829b in close_cached_tables (thd=0x7ffedc007b70, tables=0x0, wait_for_refresh=true, timeout=31536000)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:1291
#8  0x00000000016123ec in reload_acl_and_cache (thd=0x7ffedc007b70, options=16388, tables=0x0, write_to_binlog=0x7fffec5bc9dc)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_reload.cc:224
#9  0x00000000015cee9c in mysql_execute_command (thd=0x7ffedc007b70, first_level=true) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:4433
#10 0x00000000015d2fde in mysql_parse (thd=0x7ffedc007b70, parser_state=0x7fffec5bd600) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:5901
#11 0x00000000015c6b72 in dispatch_command (thd=0x7ffedc007b70, com_data=0x7fffec5bdd70, command=COM_QUERY)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:1490

(3)杀点FTWRL会话后其他select操作等待栈帧:

#0  MDL_wait::timed_wait (this=0x7ffee8008298, owner=0x7ffee8008200, abs_timeout=0x7fffec58a600, set_status_on_timeout=true, wait_state_name=0x2d897b0)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/mdl.cc:1888
#1  0x00000000016cdb30 in TABLE_SHARE::wait_for_old_version (this=0x7ffee0011620, thd=0x7ffee8008200, abstime=0x7fffec58a600, deadlock_weight=0)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/table.cc:4717
#2  0x000000000153b6ba in tdc_wait_for_old_version (thd=0x7ffee8008200, db=0x7ffee80014a0 "testmts", table_name=0x7ffee80014a8 "tii", wait_timeout=31536000, 
    deadlock_weight=0) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:2957
#3  0x000000000153ca97 in open_table (thd=0x7ffee8008200, table_list=0x7ffee8001708, ot_ctx=0x7fffec58aab0)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:3548
#4  0x000000000153f904 in open_and_process_table (thd=0x7ffee8008200, lex=0x7ffee800a830, tables=0x7ffee8001708, counter=0x7ffee800a8f0, flags=0, 
    prelocking_strategy=0x7fffec58abe0, has_prelocking_list=false, ot_ctx=0x7fffec58aab0) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:5213
#5  0x0000000001540a58 in open_tables (thd=0x7ffee8008200, start=0x7fffec58aba0, counter=0x7ffee800a8f0, flags=0, prelocking_strategy=0x7fffec58abe0)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:5831
#6  0x0000000001541e93 in open_tables_for_query (thd=0x7ffee8008200, tables=0x7ffee8001708, flags=0)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:6606
#7  0x00000000015d1dca in execute_sqlcom_select (thd=0x7ffee8008200, all_tables=0x7ffee8001708) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:5416
#8  0x00000000015ca380 in mysql_execute_command (thd=0x7ffee8008200, first_level=true) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:2939
#9  0x00000000015d2fde in mysql_parse (thd=0x7ffee8008200, parser_state=0x7fffec58c600) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:5901
#10 0x00000000015c6b72 in dispatch_command (thd=0x7ffee8008200, com_data=0x7fffec58cd70, command=COM_QUERY)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:1490

(4)占用者释放唤醒FTWRL栈帧:

Breakpoint 3, MDL_wait::set_status (this=0x7ffedc000c78, status_arg=MDL_wait::GRANTED) at /mysqldata/percona-server-locks-detail-5.7.22/sql/mdl.cc:1832
1832      bool was_occupied= TRUE;
(gdb) bt
#0  MDL_wait::set_status (this=0x7ffedc000c78, status_arg=MDL_wait::GRANTED) at /mysqldata/percona-server-locks-detail-5.7.22/sql/mdl.cc:1832
#1  0x00000000016c2483 in free_table_share (share=0x7ffee0011620) at /mysqldata/percona-server-locks-detail-5.7.22/sql/table.cc:607
#2  0x0000000001536a22 in table_def_free_entry (share=0x7ffee0011620) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:524
#3  0x00000000018fd7aa in my_hash_delete (hash=0x2e4cfe0, record=0x7ffee0011620 "\002") at /mysqldata/percona-server-locks-detail-5.7.22/mysys/hash.c:625
#4  0x0000000001537673 in release_table_share (share=0x7ffee0011620) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:949
#5  0x00000000016cad10 in closefrm (table=0x7ffee000f280, free_share=true) at /mysqldata/percona-server-locks-detail-5.7.22/sql/table.cc:3597
#6  0x0000000001537d0e in intern_close_table (table=0x7ffee000f280) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:1109
#7  0x0000000001539054 in close_thread_table (thd=0x7ffee0000c00, table_ptr=0x7ffee0000c68) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:1780
#8  0x00000000015385fe in close_open_tables (thd=0x7ffee0000c00) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:1443
#9  0x0000000001538d4a in close_thread_tables (thd=0x7ffee0000c00) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_base.cc:1722
#10 0x00000000015d19bc in mysql_execute_command (thd=0x7ffee0000c00, first_level=true) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:5307
#11 0x00000000015d2fde in mysql_parse (thd=0x7ffee0000c00, parser_state=0x7fffec5ee600) at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:5901
#12 0x00000000015c6b72 in dispatch_command (thd=0x7ffee0000c00, com_data=0x7fffec5eed70, command=COM_QUERY)
    at /mysqldata/percona-server-locks-detail-5.7.22/sql/sql_parse.cc:1490