案例如下:
在使用Show innodb status检查引擎状态时,发现了死锁问题:
*** (1) TRANSACTION:
TRANSACTION 0 677833455, ACTIVE 0 sec, process no 11393, OS thread id 278546 starting index read
mysql tables in use 1, locked 1
LOCK WAIT 3 lock struct(s), heap size 320
MySQL thread id 83, query id 162348740 dcnet03 dcnet Searching rows for update
update TSK_TASK set STATUS_ID=1064,UPDATE_TIME=now () where STATUS_ID=1061 and MON_TIME RECORD LOCKS space id 0 page no 849384 n bits 208 index `PRIMARY` of table `dcnet_db/TSK_TASK` trx id 0 677833455 lock_mode X locks rec but not gap waiting
Record lock, heap no 92 PHYSICAL RECORD: n_fields 11; compact format; info bits 0
0: len 8; hex 800000000097629c; asc b 1: len 6; hex 00002866eaee; asc (f 2: len 7; hex 00000d40040110; asc @ 3: len 8; hex 80000000000050b2; asc P 4: len 8; hex 800000000000502a; asc P*5: len 8; hex 8000000000005426; asc T&6: len 8; hex 800012412c66d29c; asc A,f 7: len 23; hex 75706c6f6164666972652e636f6d2f6 8616e642e706870; asc xxx.com/8: len 8; hex 800000000000042b; asc +9: len 4; hex 474bfa2b; asc GK +10: len 8; hex 8000000000004e24; asc N$
*** (2) TRANSACTION:
TRANSACTION 0 677833454, ACTIVE 0 sec, process no 11397, OS thread id 344086 updating or deleting, thread declared inside InnoDB 499
mysql tables in use 1, locked 1
3 lock struct(s), heap size 320, undo log entries 1
MySQL thread id 84, query id 162348739 dcnet03 dcnet Updating
update TSK_TASK set STATUS_ID=1067,UPDATE_TIME=now () where ID in (9921180)
*** (2) HOLDS THE LOCK(S):
RECORD LOCKS space id 0 page no 849384 n bits 208 index `PRIMARY` of table `dcnet_db/TSK_TASK` trx id 0 677833454 lock_mode X locks rec but not gap
Record lock, heap no 92 PHYSICAL RECORD: n_fields 11; compact format; info bits 0
0: len 8; hex 800000000097629c; asc b 1: len 6; hex 00002866eaee; asc (f 2: len 7; hex 00000d40040110; asc @ 3: len 8; hex 80000000000050b2; asc P 4: len 8; hex 800000000000502a; asc P*5: len 8; hex 8000000000005426; asc T&6: len 8; hex 800012412c66d29c; asc A,f 7: len 23; hex 75706c6f6164666972652e636f6d2f6 8616e642e706870; asc uploadfire.com/hand.php8: len 8; hex 800000000000042b; asc +9: len 4; hex 474bfa2b; asc GK +10: len 8; hex 8000000000004e24; asc N$
*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 0 page no 843102 n bits 600 index `KEY_TSKTASK_MONTIME2` of table `dcnet_db/TSK_TASK` trx id 0 677833454 lock_mode X locks rec but not gap waiting
Record lock, heap no 395 PHYSICAL RECORD: n_fields 3; compact format; info bits 0
0: len 8; hex 8000000000000425; asc %1: len 8; hex 800012412c66d29c; asc A,f 2: len 8; hex 800000000097629c; asc b
*** WE ROLL BACK TRANSACTION (1)
此死锁问题涉及TSK_TASK表,该表用于保存系统监测任务,以下是相关字段及索引:
ID:主键;
MON_TIME:监测时间;
STATUS_ID:任务状态;
索引:KEY_TSKTASK_MONTIME2 (STATUS_ID, MON_TIME)。
分析,涉及的两条语句应该不会涉及相同的TSK_TASK记录,那为什么会造成死锁呢?
查询MySQL官网文档,发现这跟MySQL的索引机制有关。MySQL的InnoDB引擎是行级锁,我原来的理解是直接对记录进行锁定,实际上并不是这样的。
要点如下:
不是对记录进行锁定,而是对索引进行锁定;
在UPDATE、DELETE操作时,MySQL不仅锁定WHERE条件扫描过的所有索引记录,而且会锁定相邻的键值,即所谓的next-key locking;
如语句UPDATE TSK_TASK SET UPDATE_TIME = NOW() WHERE ID > 10000会锁定所有主键大于等于1000的所有记录,在该语句完成之前,你就不能对主键等于10000的记录进行操作;
当非簇索引(non-cluster index)记录被锁定时,相关的簇索引(cluster index)记录也需要被锁定才能完成相应的操作。
再分析一下发生问题的两条SQL语句,就不难找到问题所在了:
当“update TSK_TASK set STATUS_ID=1064,UPDATE_TIME=now () where STATUS_ID=1061 and MON_TIME 假设“update TSK_TASK set STATUS_ID=1067,UPDATE_TIME=now () where ID in (9921180)”几乎同时执行时,本语句首先锁定簇索引(主键),由于需要更新STATUS_ID的值,所以还需要锁定KEY_TSKTASK_MONTIME2的某些索引记录。
这样第一条语句锁定了KEY_TSKTASK_MONTIME2的记录,等待主键索引,而第二条语句则锁定了主键索引记录,而等待KEY_TSKTASK_MONTIME2的记录,在此情况下,死锁就产生了。
笔者通过拆分第一条语句解决死锁问题:
先查出符合条件的ID:select ID from TSK_TASK where STATUS_ID=1061 and MON_TIME < date_sub(now(), INTERVAL 30 minute);然后再更新状态:update TSK_TASK set STATUS_ID=1064 where ID in (….)
至此,死锁问题彻底解决。