MySQL 5.7中如何定位DDL被阻塞的问题

liangkaiping0 2018-09-13

在上篇文章《MySQL表结构变更,不可不知的Metadata Lock》中,我们介绍了MDL引入的背景,及基本概念,从“道”的层面知道了什么是MDL。下面就从“术”的层面看看如何定位MDL的相关问题。

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
在配置文件中设置
 
[mysqld]
performance-schema-instrument='wait/lock/metadata/sql/mdl=ON'

测试场景

下面结合一个简单的Demo,来看看在MySQL 5.7中如何定位DDL操作的阻塞问题。

session1> begin;
Query OK, 0 rows affected (0.00 sec)

session1> delete from slowtech.t1 where id=2;
Query OK, 1 row affected (0.00 sec)

session1> select * from slowtech.t1;
+------+------+
| id  | name |
+------+------+
|    1 | a    |
+------+------+
1 row in set (0.00 sec)

session1> update slowtech.t1 set name='c' where id=1;
Query OK, 1 row affected (0.00 sec)
Rows matched: 1  Changed: 1  Warnings: 0

session2> alter table slowtech.t1 add c1 int; ##被阻塞

session3> show processlist;
+----+------+-----------+------+---------+------+---------------------------------+------------------------------------+
| Id | User | Host      | db  | Command | Time | State                          | Info                              |
+----+------+-----------+------+---------+------+---------------------------------+------------------------------------+
|  2 | root | localhost | NULL | Sleep  |  51 |                                | NULL                              |
|  3 | root | localhost | NULL | Query  |    0 | starting                        | show processlist                  |
|  4 | root | localhost | NULL | Query  |    9 | Waiting for table metadata lock | alter table slowtech.t1 add c1 int |
+----+------+-----------+------+---------+------+---------------------------------+------------------------------------+
3 rows in set (0.00 sec)

session3> select object_type,object_schema,object_name,lock_type,lock_duration,lock_status,owner_thread_id from performance_schema.metadata_locks;
+-------------+--------------------+----------------+---------------------+---------------+-------------+-----------------+
| object_type | object_schema      | object_name    | lock_type          | lock_duration | lock_status | owner_thread_id |
+-------------+--------------------+----------------+---------------------+---------------+-------------+-----------------+
| TABLE      | slowtech          | t1            | SHARED_WRITE        | TRANSACTION  | GRANTED    |              27 |
| GLOBAL      | NULL              | NULL          | INTENTION_EXCLUSIVE | STATEMENT    | GRANTED    |              29 |
| SCHEMA      | slowtech          | NULL          | INTENTION_EXCLUSIVE | TRANSACTION  | GRANTED    |              29 |
| TABLE      | slowtech          | t1            | SHARED_UPGRADABLE  | TRANSACTION  | GRANTED    |              29 |
| TABLE      | slowtech          | t1            | EXCLUSIVE          | TRANSACTION  | PENDING    |              29 |
| TABLE      | performance_schema | metadata_locks | SHARED_READ        | TRANSACTION  | GRANTED    |              28 |
+-------------+--------------------+----------------+---------------------+---------------+-------------+-----------------+
6 rows in set (0.00 sec)

这里,重点关注lock_status,"PENDING"代表线程在等待MDL,而"GRANTED"则代表线程持有MDL。

如何找出引起阻塞的会话

结合owner_thread_id,可以可到,是29号线程在等待27号线程的MDL,此时,可kill掉52号线程。

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
FROM
    performance_schema.metadata_locks a
JOIN performance_schema.metadata_locks b ON a.OBJECT_SCHEMA = b.OBJECT_SCHEMA
AND a.OBJECT_NAME = b.OBJECT_NAME
AND a.lock_status = 'PENDING'
AND b.lock_status = 'GRANTED'
AND a.OWNER_THREAD_ID <> b.OWNER_THREAD_ID
AND a.lock_type = 'EXCLUSIVE'
JOIN performance_schema.threads c ON a.OWNER_THREAD_ID = c.THREAD_ID
JOIN performance_schema.threads d ON b.OWNER_THREAD_ID = d.THREAD_ID\G

*************************** 1. row ***************************
              locked_schema: slowtech
                locked_table: t1
                locked_type: Metadata Lock
      waiting_processlist_id: 4
                waiting_age: 259
              waiting_query: alter table slowtech.t1 add c1 int
              waiting_state: Waiting for table metadata lock
    blocking_processlist_id: 2
                blocking_age: 301
              blocking_query: NULL
sql_kill_blocking_connection: KILL 2
1 row in set (0.00 sec)

输出一目了然,DDL操作如果要获得MDL,执行kill 2即可。

官方的sys.schematablelock_waits

实际上,MySQL 5.7在sys库中也集成了类似功能,同样的场景,其输出如下,

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。

但在线上,kill是一个需要谨慎的操作,毕竟你很难知道kill的是不是业务关键操作?又或者,是个批量update操作?那么,有没有办法抓到该事务之前的操作呢?

答案,有。

即Performance Schema中记录Statement Event(操作事件)的表,具体包括events_statements_current,events_statements_history,events_statements_history_long,prepared_statements_instances。

常用的是前面三个。

MySQL 5.7中如何定位DDL被阻塞的问题

三者的表结构完全一致,其中,events_statements_history又包含了events_statements_current的操作,所以我们这里会使用events_statements_history。

终极SQL如下,

SELECT
    locked_schema,
    locked_table,
    locked_type,
    waiting_processlist_id,
    waiting_age,
    waiting_query,
    waiting_state,
    blocking_processlist_id,
    blocking_age,
    substring_index(sql_text,"transaction_begin;" ,-1) AS blocking_query,
    sql_kill_blocking_connection
FROM
    (
        SELECT
            b.OWNER_THREAD_ID AS granted_thread_id,
            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
        FROM
            performance_schema.metadata_locks a
        JOIN performance_schema.metadata_locks b ON a.OBJECT_SCHEMA = b.OBJECT_SCHEMA
        AND a.OBJECT_NAME = b.OBJECT_NAME        AND a.lock_status = 'PENDING'
        AND b.lock_status = 'GRANTED'        AND a.OWNER_THREAD_ID <> b.OWNER_THREAD_ID
        AND a.lock_type = 'EXCLUSIVE'        JOIN performance_schema.threads c ON a.OWNER_THREAD_ID = c.THREAD_ID
        JOIN performance_schema.threads d ON b.OWNER_THREAD_ID = d.THREAD_ID
    ) t1,
    (
        SELECT
            thread_id,
            group_concat(  CASE WHEN EVENT_NAME = 'statement/sql/begin' THEN "transaction_begin" ELSE sql_text END ORDER BY event_id SEPARATOR ";" ) AS sql_text
        FROM
            performance_schema.events_statements_history
        GROUP BY thread_id
    ) t2
WHERE
    t1.granted_thread_id = t2.thread_id \G

*************************** 1. row ***************************
              locked_schema: slowtech
                locked_table: t1
                locked_type: Metadata Lock
      waiting_processlist_id: 4
                waiting_age: 294
              waiting_query: alter table slowtech.t1 add c1 int
              waiting_state: Waiting for table metadata lock
    blocking_processlist_id: 2
                blocking_age: 336
              blocking_query: delete from slowtech.t1 where id=2;select * from slowtech.t1;update slowtech.t1 set name='c' where id=1
sql_kill_blocking_connection: KILL 21 row in set, 1 warning (0.00 sec)

从上面的输出可以看到,blocking_query中包含了会话1中当前事务的所有操作,按执行的先后顺序输出。

需要注意的是,默认情况下,events_statements_history只会保留每个线程最近的10个操作,如果事务中进行的操作较多,实际上也是没办法抓全的。

Anyway, it is better than nothing!

相关推荐