24.锁的监控与处理浅谈

2021/7/31 23:08:25

本文主要是介绍24.锁的监控与处理浅谈,对大家解决编程问题具有一定的参考价值,需要的程序猿们随着小编来一起学习吧!

1.前言

  在Mysql中少不了与各式各样的锁打交道,比如说共享锁与排他锁等等,其中有时我们可能会遇到锁等待的情况,比如说同时开启两个事务对某一行记录进行修改,其中一个事务修改完但是未提交,另一个事务也修改,那么此事它就会遇到锁等待的情况(就是阻塞-->block)

2. show status like 'innodb_row_lock%' 

root@localhost 17:19:  [liulin]> show status like 'innodb_row_lock%';
+-------------------------------+-------+
| Variable_name                 | Value |
+-------------------------------+-------+
| Innodb_row_lock_current_waits | 1     |             # 当前处于锁等待的个数
| Innodb_row_lock_time          | 51068 |        #InnoDB 表等待获取行锁的总时间开销,单位毫秒,
| Innodb_row_lock_time_avg      | 25534 |
| Innodb_row_lock_time_max      | 51068 |
| Innodb_row_lock_waits         | 2     |             # 这个表示数据库从启动开始直到现在共发生锁等待的次数(包括目前正在处于锁等待的个数)
+-------------------------------+-------+

3. select * from information_schema.innodb_trx where trx_state='lock wait'\G;  

root@localhost 20:28:  [information_schema]> select * from information_schema.innodb_trx where trx_state='lock wait'\G;
*************************** 1. row ***************************
                    trx_id: 35182
                 trx_state: LOCK WAIT
               trx_started: 2021-07-31 17:18:11
     trx_requested_lock_id: 35182:116:3:11
          trx_wait_started: 2021-07-31 20:27:29
                trx_weight: 2
       trx_mysql_thread_id: 37
                 trx_query: update t1 set name='liulin1' where id=1
       trx_operation_state: starting index read
         trx_tables_in_use: 1
         trx_tables_locked: 1
          trx_lock_structs: 2
     trx_lock_memory_bytes: 1136
           trx_rows_locked: 4
         trx_rows_modified: 0
   trx_concurrency_tickets: 0
       trx_isolation_level: REPEATABLE READ
         trx_unique_checks: 1
    trx_foreign_key_checks: 1
trx_last_foreign_key_error: NULL
 trx_adaptive_hash_latched: 0
 trx_adaptive_hash_timeout: 0
          trx_is_read_only: 0
trx_autocommit_non_locking: 0
1 row in set (0.00 sec)
  • 这个表示查找处于锁等待的事务的详细信息,通俗的说是它目前想要获取写锁,但是该锁被被的事务提前一步获取且还未释放,因此它想要获取就处于了阻塞状态了(blocked).
  • 一般这里我们我们会设置一下锁超时等待时间的,参数为innodb_lock_wait_timeout,默认是50s,表示innoDB 事务请求行锁的超时时间限制参数
  • select * from information_schema.innodb_trx  :可以查找出当前数据库中所有未提交的事务

4. select * from sys.innodb_lock_waits;   查看被锁事务和锁源事务的信息

root@localhost 20:48:  [sys]> select * from innodb_lock_waits \G;
*************************** 1. row ***************************
                wait_started: 2021-07-31 20:48:26
                    wait_age: 00:00:08
               wait_age_secs: 8
                locked_table: `liulin`.`t1`
                locked_index: PRIMARY
                 locked_type: RECORD
              waiting_trx_id: 35182
         waiting_trx_started: 2021-07-31 17:18:11
             waiting_trx_age: 03:30:23
     waiting_trx_rows_locked: 7
   waiting_trx_rows_modified: 0
                 waiting_pid: 37
               waiting_query: update t1 set name='liulin1' where id=1
             waiting_lock_id: 35182:116:3:11
           waiting_lock_mode: X
             blocking_trx_id: 35181
                blocking_pid: 36
              blocking_query: NULL
            blocking_lock_id: 35181:116:3:11
          blocking_lock_mode: X
        blocking_trx_started: 2021-07-31 17:17:12
            blocking_trx_age: 03:31:22
    blocking_trx_rows_locked: 1
  blocking_trx_rows_modified: 1
     sql_kill_blocking_query: KILL QUERY 36
sql_kill_blocking_connection: KILL 36
1 row in set, 3 warnings (0.00 sec)

其中:waiting_trx_id:被锁的事务id   waiting_pid :被锁的线程id    

   blocking_trx_id:锁源的事务id   blocking_pid :表示锁源的线程id

5. select * from performance_schema.threads where processlist_id='xxx' 

root@localhost 21:01:  [performance_schema]> select * from threads  where processlist_id=36\G;
*************************** 1. row ***************************
          THREAD_ID: 62
               NAME: thread/sql/one_connection
               TYPE: FOREGROUND
     PROCESSLIST_ID: 36
   PROCESSLIST_USER: root
   PROCESSLIST_HOST: localhost
     PROCESSLIST_DB: liulin
PROCESSLIST_COMMAND: Sleep
   PROCESSLIST_TIME: 13441
  PROCESSLIST_STATE: NULL
   PROCESSLIST_INFO: update t1 set name='liulin1' where id=1
   PARENT_THREAD_ID: NULL
               ROLE: NULL
       INSTRUMENTED: YES
            HISTORY: YES
    CONNECTION_TYPE: Socket
       THREAD_OS_ID: 4548
1 row in set (0.00 sec)

6.  select *  from  performance_schema.events_statements_current where thread_id='xxx';

root@localhost 22:24:  [performance_schema]> select *  from  performance_schema.events_statements_current  where thread_id='62'\G;
*************************** 1. row ***************************
              THREAD_ID: 62
               EVENT_ID: 46
           END_EVENT_ID: 46
             EVENT_NAME: statement/sql/update
                 SOURCE: 
            TIMER_START: 7601338054145000
              TIMER_END: 7601339393285000
             TIMER_WAIT: 1339140000
              LOCK_TIME: 277000000
               SQL_TEXT: update t1 set name='liulin1' where id=1
                 DIGEST: e63566e2378e6dbf044cfe5345a79a9a
            DIGEST_TEXT: UPDATE `t1` SET NAME = ? WHERE `id` = ? 
         CURRENT_SCHEMA: liulin
            OBJECT_TYPE: NULL
          OBJECT_SCHEMA: NULL
            OBJECT_NAME: NULL
  OBJECT_INSTANCE_BEGIN: NULL
            MYSQL_ERRNO: 0
      RETURNED_SQLSTATE: 00000
           MESSAGE_TEXT: Rows matched: 1  Changed: 1  Warnings: 0
                 ERRORS: 0
               WARNINGS: 0
          ROWS_AFFECTED: 1
              ROWS_SENT: 0
          ROWS_EXAMINED: 1
CREATED_TMP_DISK_TABLES: 0
     CREATED_TMP_TABLES: 0
       SELECT_FULL_JOIN: 0
 SELECT_FULL_RANGE_JOIN: 0
           SELECT_RANGE: 0
     SELECT_RANGE_CHECK: 0
            SELECT_SCAN: 0
      SORT_MERGE_PASSES: 0
             SORT_RANGE: 0
              SORT_ROWS: 0
              SORT_SCAN: 0
          NO_INDEX_USED: 0
     NO_GOOD_INDEX_USED: 0
       NESTING_EVENT_ID: NULL
     NESTING_EVENT_TYPE: NULL
    NESTING_EVENT_LEVEL: 0
1 row in set (0.00 sec)

 

总结:

  show status like 'innodb_row_lock%'   ---> select *  from information_schema.innodb.trx where trx_state='lock wait' -->

  select * from sys.innodb_lock_waits    ---> select * from performance_schema.threads  where processlist_id='xxx'  --->

  select * from performance_schema.events_statements_current where thread_id='xxx';

  其中,第二个命令如果是只是查看锁等待的事务的详细信息的话,我们可以直接用第三个命令代替,因为第二个命令还可以查看所有状态的事务的详细信息。然后第三个命令也是查看锁源事务和被锁事务的一些详细信息,然后第四个命令可以针对连接线程id查看其内部的Mysql线程id,最后第五个命令最后再根据第四个查到的Mysql内部的线程id查看到该线程id对应的当前线程id中对应的事件。

  

案例篇:

  背景:

    硬件环境:DELL R720 , 16核,48G,SAS 900G, RAID10

    在巡检的时候发现,在某个时间段的cpu压力非常高

  排查和处理过程

    1.通过top命令查看,发现Mysqld进程占比非常高

    2.用vmstat查看发现大量的CPU是被用作了sys和wait,us处于正常

    3,这时怀疑是Mysql的锁或者sql语句出现了问题

    4.通过show status like 'innodb_row_lock%'查看有大量的所等待以及通过查看慢日志发现有少量的慢语句

    5.其中关于慢日志可以通过pt-query-disget工具进行分析 

 

 

 

 

 



这篇关于24.锁的监控与处理浅谈的文章就介绍到这儿,希望我们推荐的文章对大家有所帮助,也希望大家多多支持为之网!


扫一扫关注最新编程教程