有时候,怀疑某个MySQL内存查询导致CPU或磁盘I/O消耗特别高,但又不确定具体是哪个SQL引起的。
或者当InnoDB引擎内部有semaphore wait
时,想知道具体是哪个线程/查询引起的。多说一下,当有semaphore wait
事件超过600秒的话,InnoDB会发出crash信号:
InnoDB: ###### Diagnostic info printed to the standard error stream 2020-12-13T09:41:33.810011Z 0 [ERROR] [FATAL] InnoDB: Semaphore wait has lasted > 600 seconds. We intentionally crash the server because it appears to be hung. 2020-12-13 10:41:33 0x7f3d92a4e700 InnoDB: Assertion failure in thread 139902430013184 in file ut0ut.cc line 917 InnoDB: We intentionally generate a memory trap. InnoDB: Submit a detailed bug report to http://bugs.mysql.com. InnoDB: If you get repeated assertion failures or crashes, even InnoDB: immediately after the mysqld startup, there may be InnoDB: corruption in the InnoDB tablespace. Please refer to InnoDB: http://dev.mysql.com/doc/refman/8.0/en/forcing-innodb-recovery.html InnoDB: about forcing recovery. 09:41:33 UTC - mysqld got signal 6 ;
因此也要监控InnoDB的semaphore wait
状态,一旦超过阈值,就要尽快报警并分析出问题原因,及时杀掉或停止引起等待的查询请求。
MySQL是一个单进程多线程的服务程序,用 ps -ef|grep mysqld
就能看到其系统进程ID了。另外,当 my.cnf
配置文件中增加一行 innodb_status_file = 1
时,也会生成带有系统进程ID的innodb status 文件
[root@yejr.run]# ps -ef | grep mysqld mysql 38801 1 0 Jun13 ? 00:03:30 /usr/local/GreatSQL-8.0.22/bin/mysqld --defaults-file=/mysql/data06/my.cnf [root@yejr.run]# ls -la innodb_status.38801 -rw-r----- 1 mysql mysql 4906 Jun 14 14:26 innodb_status.38801
文件 innodb_status.pid
的作用是每隔15秒左右输出innodb引擎各种状态信息,和执行 SHOW ENGINE INNODB STATUS
的作用相同。二者的区别在于,前者(文件输出方式)的输出内容长度不受限制,而后者(命令行输出)则最多只显示1MB内容,更多的会被截断。所以务必设置 innodb_status_file = 1
选项。
Standard Monitor output is limited to 1MB when produced using the SHOW ENGINE INNODB STATUS statement. This limit does not apply to output written to server standard error output (stderr).
从MySQL 5.7开始,performance_schema.threads
表增加 THREAD_OS_ID
列,用于记录MySQL内部线程对应的系统线程ID。
创建一个新连接,并执行下面的SQL:
[root@yejr.run]# mysql -S./mysql.sock -uroot -p mymgr ... # 查到MySQL的连接ID是25(PROCESSLIST_ID = 25) [root@yejr.run] [none]> show processlist; +----+------+-----------+-------+---------+------+-------+------------------+-----------+---------------+ | Id | User | Host | db | Command | Time | State | Info | Rows_sent | Rows_examined | +----+------+-----------+-------+---------+------+-------+------------------+-----------+---------------+ | 25 | root | localhost | mymgr | Query | 0 | init | show processlist | 0 | 0 | +----+------+-----------+-------+---------+------+-------+------------------+-----------+---------------+ [root@yejr.run] [mymgr]> begin; select *,sleep(1000) from t1 for update; ... <-- 这个SQL会运行很长时间,方便我们观察
新开一个窗口,查看 pfs.threads
表:
[root@yejr.run] [performance_schema]> SELECT * FROM threads WHERE PROCESSLIST_ID=25\G *************************** 1. row *************************** THREAD_ID: 65 <-- MySQL内部线程ID,也是PFS的内部计数器 NAME: thread/sql/one_connection TYPE: FOREGROUND PROCESSLIST_ID: 25 <-- MySQL连接ID PROCESSLIST_USER: root PROCESSLIST_HOST: localhost PROCESSLIST_DB: mymgr PROCESSLIST_COMMAND: Query PROCESSLIST_TIME: 246 PROCESSLIST_STATE: User sleep PROCESSLIST_INFO: select *,sleep(1000) from t1 for update <-- 正在运行的SQL PARENT_THREAD_ID: NULL ROLE: NULL INSTRUMENTED: YES HISTORY: YES CONNECTION_TYPE: Socket THREAD_OS_ID: 58412 <-- 对应操作系统的线程ID RESOURCE_GROUP: USR_default
运行 ps -Lef 查看对应的系统线程
[root@yejr.run]# ps -Lef | grep 58412 <-- 上面查询pfs.threads看到 THREAD_OS_ID 列的值 mysql 38801 1 58412 0 40 14:46 ? 00:00:00 /usr/local/GreatSQL-8.0.22/bin/mysqld --defaults-file=/mysql/data06/my.cnf
如果设置了 general_log=1
,那么也能看到 general_log
里有这样的记录:
[root@yejr.run]# cat yejr.run.log /usr/local/GreatSQL-8.0.22/bin/mysqld, Version: 8.0.22-13 (Source distribution). started with: Tcp port: 6001 Unix socket: mysql.sock #几个列分别是:时间、连接ID、请求类型、详细的SQL Time Id Command Argument ... 2021-06-14T14:46:47.474393+08:00 25 Query select *,sleep(1000) from t1 for update <-- 可以看到连接ID是25 ...
再查询 pfs.events_statements_current
表:
[root@yejr.run] [performance_schema]> SELECT * FROM events_statements_current WHERE THREAD_ID = 65\G THREAD_ID: 65 EVENT_ID: 8 END_EVENT_ID: NULL EVENT_NAME: statement/sql/select SOURCE: init_net_server_extension.cc:95 TIMER_START: 82217983305961000 TIMER_END: 82559992238886000 TIMER_WAIT: 342008932925000 LOCK_TIME: 206000000 SQL_TEXT: select *,sleep(1000) from t1 for update DIGEST: 4d7f4182dff4abc484010b73024b4afb35075e1b6592d14ce895a2b8764b8f46 DIGEST_TEXT: SELECT * , `sleep` (?) FROM `t1` FOR UPDATE CURRENT_SCHEMA: mymgr ... SELECT_SCAN: 1 ... NESTING_EVENT_ID: 6 NESTING_EVENT_TYPE: TRANSACTION NESTING_EVENT_LEVEL: 0 STATEMENT_ID: 87 <-- 查询ID,每次查询该ID值都会增加
执行 SHOW ENGINE INNODB STATUS\G
查看事务状态:
... # 事务ID=9322,运行时长=252秒 ---TRANSACTION 9322, ACTIVE 252 sec mysql tables in use 1, locked 1 2 lock struct(s), heap size 1136, 1 row lock(s) # MySQL连接ID=25,OS线程句柄 = 140442126739200(后面再介绍),查询ID=87(对应上面的 STATEMENT_ID) MySQL thread id 25, OS thread handle 140442126739200, query id 87 localhost root User sleep select *,sleep(1000) from t1 for update ...
首先,OS thread handle 140442126739200
(OS thread handle是进程内部用于识别各个线程的内部ID),这里是个十进制的数值,需要先转成十六进制(有时候可能会直接用十六进制表示,这个问题也有人提出了 MDEV-17237):
[root@yejr.run] [performance_schema]> select lower(conv(140442126739200, 10, 16)); +--------------------------------------+ | lower(conv(140442126739200, 10, 16)) | +--------------------------------------+ | 7fbb3b136700 | +--------------------------------------+
再利用 pstack 查询该句柄和操作系统线程ID的关联:
[root@yejr.run]# pstack `pidof mysqld` | grep 7fbb3b136700 Thread 2 (Thread 0x7fbb3b136700 (LWP 58412)):
可以看到 LWP = 58412
,对应上面的 THREAD_OS_ID
值,LWP是Light-Weight Processes的缩写(轻量级进程)。用 pidstat 也能看到这个LWP:
[root@yejr.run]# pidstat -t -p 38801 | grep 58031 03:45:02 PM 1000 - 58031 0.00 0.00 0.00 0.00 2 |__mysqld
【特别提醒】运行pstack会短暂阻塞mysqld进程,所以请切勿在业务高峰期执行,除非万不得已。
有时候可能会看到类似下面的 innodb status:
SEMAPHORES ---------- OS WAIT ARRAY INFO: reservation count 15143 --Thread 140585128785664 has waited at buf0flu.cc line 1209 for 237 seconds the semaphore: SX-lock on RW-latch at 0x7fdb1fbe3f80 created in file buf0buf.cc line 1460 a writer (thread id 140584786024192) has reserved it in mode SX number of readers 0, waiters flag 1, lock_word: 10000000 Last time read locked in file row0sel.cc line 3758 Last time write locked in file /export/home/pb2/build/sb_0-34537258-1560179931.8/mysql-5.7.27/storage/innobase/fsp/fsp0fsp.cc line 167
在上面这段信息中,线程 Thread 140585128785664
在 buf0flu.cc
代码1209行这里等待了237秒,想要请求 SX-lock 的 RW-latch,被另一个线程 thread id 140584786024192
给阻塞了,它持有的是 SX latch。
这时候,就可以利用pstack反查对应的LWP,再反查出PROCESSLIST_ID,以及正在执行的SQL。pfs.events_statements_current
中只能查到当前执行的SQL,可以通过 pfs.events_statements_history
查看最近执行过的SQL。
Enjoy MySQL :)