首页 > 数据库 > MySQL > 正文

MySQL 5.7中如何定位DDL被阻塞的难题

2024-07-24 12:31:48
字体:
来源:转载
供稿:网友
        在MySQL 5.7中,针对MDL,引入了一张新表performance_schema.metadata_locks,该表可对外展示MDL的相关信息,包括其作用对象,类型及持有等待情况。
 
        开启MDL的instrument
        但是相关instrument并没有开启(MySQL 8.0是默认开启的),其可通过如下两种方式开启,
 
       临时生效
 
       修改performance_schema.setup_instrume nts表,但实例重启后,又会恢复为默认值。
   
       UPDATE performance_schema.setup_instruments SET ENABLED = 'YES', TIMED = 'YES'WHERE NAME = 'wait/lock/metadata/sql/mdl';
        永久生效
 
但需要注意的是,owner_thread_id给出的只是线程ID,并不是show processlist中的ID。如果要查找线程对应的processlist id,需查询performance_schema.threads表。
 
session3> select * from performance_schema.threads where thread_id in (27,29)/G
*************************** 1. row ***************************
          THREAD_ID: 27
               NAME: thread/sql/one_connection
               TYPE: FOREGROUND
     PROCESSLIST_ID: 2
   PROCESSLIST_USER: root
   PROCESSLIST_HOST: localhost
     PROCESSLIST_DB: NULL
PROCESSLIST_COMMAND: Sleep
   PROCESSLIST_TIME: 214
  PROCESSLIST_STATE: NULL
   PROCESSLIST_INFO: NULL
   PARENT_THREAD_ID: 1
               ROLE: NULL
       INSTRUMENTED: YES
            HISTORY: YES
    CONNECTION_TYPE: Socket
       THREAD_OS_ID: 9800
*************************** 2. row ***************************
          THREAD_ID: 29
               NAME: thread/sql/one_connection
               TYPE: FOREGROUND
     PROCESSLIST_ID: 4
   PROCESSLIST_USER: root
   PROCESSLIST_HOST: localhost
     PROCESSLIST_DB: NULL
PROCESSLIST_COMMAND: Query
   PROCESSLIST_TIME: 172
  PROCESSLIST_STATE: Waiting for table metadata lock
   PROCESSLIST_INFO: alter table slowtech.t1 add c1 int
   PARENT_THREAD_ID: 1
               ROLE: NULL
       INSTRUMENTED: YES
            HISTORY: YES
    CONNECTION_TYPE: Socket
       THREAD_OS_ID: 9907
2 rows in set (0.00 sec)
将这两张表结合,借鉴sys.innodb_lock _waits的输出,实际上我们也可以直观地呈现MDL的等待关系。
 
SELECT
    a.OBJECT_SCHEMA AS locked_schema,
    a.OBJECT_NAME AS locked_table,
    "Metadata Lock" AS locked_type,
    c.PROCESSLIST_ID AS waiting_processlist_id,
    c.PROCESSLIST_TIME AS waiting_age,
    c.PROCESSLIST_INFO AS waiting_query,
    c.PROCESSLIST_STATE AS waiting_state,
    d.PROCESSLIST_ID AS blocking_processlist_id,
    d.PROCESSLIST_TIME AS blocking_age,
    d.PROCESSLIST_INFO AS blocking_query,
    concat('KILL ', d.PROCESSLIST_ID) AS sql_kill_blocking_connection
 
mysql> select * from sys.schema_table_lock_waits/G
*************************** 1. row ***************************
               object_schema: slowtech
                 object_name: t1
           waiting_thread_id: 29
                 waiting_pid: 4
             waiting_account: root@localhost
           waiting_lock_type: EXCLUSIVE
       waiting_lock_duration: TRANSACTION
               waiting_query: alter table slowtech.t1 add c1 int
          waiting_query_secs: 446
 waiting_query_rows_affected: 0
 waiting_query_rows_examined: 0
          blocking_thread_id: 27
                blocking_pid: 2
            blocking_account: root@localhost
          blocking_lock_type: SHARED_READ
      blocking_lock_duration: TRANSACTION
     sql_kill_blocking_query: KILL QUERY 2
sql_kill_blocking_connection: KILL 2
*************************** 2. row ***************************
               object_schema: slowtech
                 object_name: t1
           waiting_thread_id: 29
                 waiting_pid: 4
             waiting_account: root@localhost
           waiting_lock_type: EXCLUSIVE
       waiting_lock_duration: TRANSACTION
               waiting_query: alter table slowtech.t1 add c1 int
          waiting_query_secs: 446
 waiting_query_rows_affected: 0
 waiting_query_rows_examined: 0
          blocking_thread_id: 29
                blocking_pid: 4
            blocking_account: root@localhost
          blocking_lock_type: SHARED_UPGRADABLE
      blocking_lock_duration: TRANSACTION
     sql_kill_blocking_query: KILL QUERY 4
sql_kill_blocking_connection: KILL 4
2 rows in set (0.00 sec)
具体分析下官方的输出,
 
只有一个alter table操作,却产生了两条记录,而且两条记录的kill对象竟然还不一样,对表结构不熟悉及不仔细看记录内容的话,难免会kill错对象。
 
不仅如此,如果有N个查询被DDL操作堵塞,则会产生N*2条记录。在阻塞操作较多的情况下,这N*2条记录完全是个噪音。
 
而之前的SQL,无论有多少操作被阻塞,一个alter table操作,就只会输出一条记录。
 
如何查看阻塞会话已经执行过的操作
但上面这个SQL也有遗憾,其blocking_query为NULL,而在会话1中,其明明已经执行了三个SQL。
 
这个与performance_schema.threads(类似于show processlist)有关,其只会输出当前正在运行的SQL,对于已经执行过的,实际上是没办法看到。

(编辑:武林网)

发表评论 共有条评论
用户名: 密码:
验证码: 匿名发表