背景
以前接触到的数据库死锁,都是批量更新时加锁顺序不一致而导致的死锁,但是上周却遇到了一个很难理解的死锁。借着这个机会又重新学习了一下mysql的死锁知识以及常见的死锁场景。在多方调研以及和同事们的讨论下终于发现了这个死锁问题的成因,收获颇多。虽然是后端程序员,我们不需要像DBA一样深入地去分析与锁相关的源码,但是如果我们能够掌握基本的死锁排查方法,对我们的日常开发还是大有裨益的。
PS:本文不会介绍死锁的基本知识,mysql的加锁原理可以参考本文的参考资料提供的链接。
死锁起因
先介绍一下数据库和表情况,因为涉及到公司内部真是的数据,所以以下都做了模拟,不会影响具体的分析。
我们采用的是5.5版本的mysql数据库,事务隔离级别是默认的RR(Repeatable-Read),采用innodb引擎。假设存在test表:
CREATE TABLE `test` ( `id` int(11) unsigned NOT NULL AUTO_INCREMENT, `a` int(11) unsigned DEFAULT NULL, PRIMARY KEY (`id`), UNIQUE KEY `a` (`a`) ) ENGINE=InnoDB AUTO_INCREMENT=100 DEFAULT CHARSET=utf8; |
表的结构很简单,一个主键id,另一个唯一索引a。表里的数据如下:
mysql> select * from test; +----+------+ | id | a | +----+------+ | 1 | 1 | | 2 | 2 | | 4 | 4 | +----+------+ 3 rows in set (0.00 sec) |
出现死锁的操作如下:
然后我们可以通过SHOW ENGINE INNODB STATUS;来查看死锁日志:
------------------------ LATEST DETECTED DEADLOCK ------------------------ 170219 13:31:31 *** (1) TRANSACTION: TRANSACTION 2A8BD, ACTIVE 11 sec starting index read mysql tables in use 1, locked 1 LOCK WAIT 2 lock struct(s), heap size 376, 1 row lock(s) MySQL thread id 448218, OS thread handle 0x2abe5fb5d700, query id 18923238 renjun.fangcloud.net 121.41.41.92 root updating delete from test where a = 2 *** (1) WAITING FOR THIS LOCK TO BE GRANTED: RECORD LOCKS space id 0 page no 923 n bits 80 index `a` of table `oauthdemo`.`test` trx id 2A8BD lock_mode X waiting Record lock, heap no 3 PHYSICAL RECORD: n_fields 2; compact format; info bits 32 0: len 4; hex 00000002; asc ;; 1: len 4; hex 00000002; asc ;; *** (2) TRANSACTION: TRANSACTION 2A8BC, ACTIVE 18 sec inserting mysql tables in use 1, locked 1 4 lock struct(s), heap size 1248, 3 row lock(s), undo log entries 2 MySQL thread id 448217, OS thread handle 0x2abe5fd65700, query id 18923239 renjun.fangcloud.net 121.41.41.92 root update insert into test (id,a) values (10,2) *** (2) HOLDS THE LOCK(S): RECORD LOCKS space id 0 page no 923 n bits 80 index `a` of table `oauthdemo`.`test` trx id 2A8BC lock_mode X locks rec but not gap Record lock, heap no 3 PHYSICAL RECORD: n_fields 2; compact format; info bits 32 0: len 4; hex 00000002; asc ;; 1: len 4; hex 00000002; asc ;; *** (2) WAITING FOR THIS LOCK TO BE GRANTED: RECORD LOCKS space id 0 page no 923 n bits 80 index `a` of table `oauthdemo`.`test` trx id 2A8BC lock mode S waiting Record lock, heap no 3 PHYSICAL RECORD: n_fields 2; compact format; info bits 32 0: len 4; hex 00000002; asc ;; 1: len 4; hex 00000002; asc ;; *** WE ROLL BACK TRANSACTION (1) |
分析
阅读死锁日志
遇到死锁,第一步就是阅读死锁日志。死锁日志通常分为两部分,上半部分说明了事务1在等待什么锁:
170219 13:31:31 *** (1) TRANSACTION: TRANSACTION 2A8BD, ACTIVE 11 sec starting index read mysql tables in use 1, locked 1 LOCK WAIT 2 lock struct(s), heap size 376, 1 row lock(s) MySQL thread id 448218, OS thread handle 0x2abe5fb5d700, query id 18923238 renjun.fangcloud.net 121.41.41.92 root updating delete from test where a = 2 *** (1) WAITING FOR THIS LOCK TO BE GRANTED: RECORD LOCKS space id 0 page no 923 n bits 80 index `a` of table `oauthdemo`.`test` trx id 2A8BD lock_mode X waiting Record lock, heap no 3 PHYSICAL RECORD: n_fields 2; compact format; info bits 32 0: len 4; hex 00000002; asc ;; 1: len 4; hex 00000002; asc ;; |
从日志里我们可以看到事务1当前正在执行delete from test where a = 2,该条语句正在申请索引a的X锁,所以提示lock_mode X waiting。
然后日志的下半部分说明了事务2当前持有的锁以及等待的锁:
*** (2) TRANSACTION: TRANSACTION 2A8BC, ACTIVE 18 sec inserting mysql tables in use 1, locked 1 4 lock struct(s), heap size 1248, 3 row lock(s), undo log entries 2 MySQL thread id 448217, OS thread handle 0x2abe5fd65700, query id 18923239 renjun.fangcloud.net 121.41.41.92 root update insert into test (id,a) values (10,2) *** (2) HOLDS THE LOCK(S): RECORD LOCKS space id 0 page no 923 n bits 80 index `a` of table `oauthdemo`.`test` trx id 2A8BC lock_mode X locks rec but not gap Record lock, heap no 3 PHYSICAL RECORD: n_fields 2; compact format; info bits 32 0: len 4; hex 00000002; asc ;; 1: len 4; hex 00000002; asc ;; *** (2) WAITING FOR THIS LOCK TO BE GRANTED: RECORD LOCKS space id 0 page no 923 n bits 80 index `a` of table `oauthdemo`.`test` trx id 2A8BC lock mode S waiting Record lock, heap no 3 PHYSICAL RECORD: n_fields 2; compact format; info bits 32 0: len 4; hex 00000002; asc ;; 1: len 4; hex 00000002; asc ;; |
从日志的HOLDS THE LOCKS(S)块中我们可以看到事务2持有索引a的X锁,并且是记录锁(Record Lock)。该锁是通过事务2在步骤2执行的delete语句申请的。由于是RR隔离模式下的基于唯一索引的等值查询(Where a = 2),所以会申请一个记录锁,而非next-key锁。