Innodb主键和辅助索引的更新锁顺序

Question

当我研究由两个更新查询引起的死锁时。这是有道理的，我不明白。1.行锁设置顺序? 2.主索引和辅助索引上的锁设置顺序，当更新执行时? 3.显示INNODB状态显示等待的x锁结构，它是同时需要的还是在其他授予之后需要的? 4.如果一个锁结构不想锁某些行，进程是原子的吗？

下面是我的死锁:表：

CREATE TABLE `study_update_deadlock` (
    `id` int(11) NOT NULL AUTO_INCREMENT,
    `u_key` bigint(20) DEFAULT NULL,
    `nu_key` bigint(20) DEFAULT NULL,
    `version` int(11) DEFAULT NULL,
    `quantity` int(11) DEFAULT NULL,
    PRIMARY KEY (`id`),
    KEY `nu_key` (`nu_key`),
    KEY `u_key` (`u_key`)
) ENGINE=InnoDB AUTO_INCREMENT=5000001 DEFAULT CHARSET=latin1 COMMENT='根据非唯一主键更新是发生死锁'

更新查询1：

update study_update_deadlock set version=version+1 where u_key=1663577608119220637 and nu_key=12498159

更新查询2：

update study_update_deadlock set quantity=quantity+1 where u_key=1470344318505049187 and nu_key=12498159

DB ExampleRows中的若干行

方案的关键：

1. 根据nu_key索引更新
2. 两个查询更新不同的行，但两行具有相同的nu_key索引
3. TRX1锁等待5锁结构？是否将所有锁要求添加到锁图中？
4. TRX2持有PrimaryKey，并等待锁定nu_key索引。但是根据我所知道的先设置索引锁，锁要求不是原子的吗？
5. 隔离等级:RR

DeadLock显示INNODB状态

2016-06-29 18:58:32 700001f3b000
*** (1) TRANSACTION:
TRANSACTION 76027, ACTIVE 0 sec fetching rows
mysql tables in use 3, locked 3
LOCK WAIT 5 lock struct(s), heap size 1184, 8 row lock(s)
MySQL thread id 33311, OS thread handle 0x700001ab7000, query id 204129 localhost root updating
update study_update_deadlock set quantity = quantity+1 where u_key= 1470344318505049187 and nu_key=12498159
*** (1) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 16 page no 22255 n bits 392 index `PRIMARY` of table `test`.`study_update_deadlock` trx id 76027 lock_mode X locks rec but not gap waiting
Record lock, heap no 255 PHYSICAL RECORD: n_fields 7; compact format; info bits 0
 0: len 4; hex 8036731e; asc  6s ;;
 1: len 6; hex 0000000128fa; asc     ( ;;
 2: len 7; hex 770000179e081e; asc w      ;;
 3: len 8; hex 97163705443d799d; asc   7 D=y ;;
 4: len 8; hex 8000000000beb4ef; asc         ;;
 5: len 4; hex 800000bb; asc     ;;
 6: len 4; hex 8000005a; asc    Z;;

*** (2) TRANSACTION:
TRANSACTION 76028, ACTIVE 0 sec starting index read
mysql tables in use 3, locked 3
4 lock struct(s), heap size 1184, 3 row lock(s)
MySQL thread id 33312, OS thread handle 0x700001f3b000, query id 204130 localhost root updating
update study_update_deadlock set version = version+1 where u_key= 1663577608119220637 and nu_key=12498159
*** (2) HOLDS THE LOCK(S):
RECORD LOCKS space id 16 page no 22255 n bits 392 index `PRIMARY` of table `test`.`study_update_deadlock` trx id 76028 lock_mode X locks rec but not gap
Record lock, heap no 255 PHYSICAL RECORD: n_fields 7; compact format; info bits 0
 0: len 4; hex 8036731e; asc  6s ;;
 1: len 6; hex 0000000128fa; asc     ( ;;
 2: len 7; hex 770000179e081e; asc w      ;;
 3: len 8; hex 97163705443d799d; asc   7 D=y ;;
 4: len 8; hex 8000000000beb4ef; asc         ;;
 5: len 4; hex 800000bb; asc     ;;
 6: len 4; hex 8000005a; asc    Z;;

*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 16 page no 22336 n bits 952 index `nu_key` of table `test`.`study_update_deadlock` trx id 76028 lock_mode X waiting
Record lock, heap no 660 PHYSICAL RECORD: n_fields 2; compact format; info bits 0
 0: len 8; hex 8000000000beb4ef; asc         ;;
 1: len 4; hex 8036731c; asc  6s ;;

*** WE ROLL BACK TRANSACTION (2)

Answer 1

这是由间隙锁和记录锁造成的。GDB欺骗mysql源代码可以发现，MySQL内部DOC有一些KeyPoint，但丢失了一些细节。