Next-key locking是如何解决幻读问题的
首先什么是幻读呢？
举个例子，两个男孩同时在追求一个女生的故事
A问：你有男朋友吗？女孩对他说没有。A追求女孩的事件还没有提交，就是继续追求哈。
就在A追求的同时，B也在追求，并且直接让女孩做他的女朋友，女孩答应了，B的追求事件结束。
A又问：你有男朋友吗？ 女孩对他说我已经有男朋友了！ 呜呜呜 ！刚才你还没有的，怎么现在就有了呢？
女孩说，你也没说过你追我的时候不让别人追我啊！... ... A哭着走了。
幻读 Phantom Problem 是指在同一事务下，连续执行两次相同的sql语句可能导致不同的结果，第二次的sql语句可能会返回之前不存在的行。
在刚才我举的例子里，A虽然问了女孩有没有男朋友，但是没有告诉女孩，在他追求时，不可以接受别人的追求，所以悲催的结局。
那么A怎么才能在他追求事件结束前让女孩不答应别人的追求呢？
innodb中的RR隔离级别是通过next-key locking是如何解决幻读问题的，就是锁住一个范围。
那么如果你是A你怎么做呢？你肯定要跟女孩说，只要我开始追求你，问了你有没有男朋友，在我结束追求你之前，你不可以答应别人的追求！我要把你脑子里记录男朋友的区域全部锁起来，啊哈啊！
下面我们来做一个测试，分别在RR和RC隔离级别中来实现：
测试使用表db1.t1 (a int primary key) ,记录有1,3,5
| T1 RC | T2 RR |
|---|
| begin; | begin; |
| set session transaction isolation level READ COMMITTED; | |
| select * from db1.t1 where a>3 for update; | |
| 查询结果为5 | |
| | insert into db1.t1 values (4); |
| | commit; |
| select * from db1.t1 where a>3; | |
| 查询结果为4 5 | |
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 | MariaDB [db1]> create table t1 (a int primary key);
Query OK, 0 rows affected (0.22 sec)
MariaDB [db1]> insert into t1 values (1),(3),(5);
Query OK, 3 rows affected (0.02 sec)
Records: 3 Duplicates: 0 Warnings: 0
#事务T1
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> set session transaction isolation level read co
Query OK, 0 rows affected (0.01 sec)
MariaDB [db1]> select * from db1.t1 where a>3 for update;
+---+
| a |
+---+
| 5 |
+---+
1 row in set (0.01 sec)
#事务T2
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> insert into db1.t1 values (4);
Query OK, 1 row affected (0.00 sec)
MariaDB [db1]> commit;
Query OK, 0 rows affected (0.03 sec)
#事务T1
MariaDB [db1]> select * from db1.t1 where a>3 for update;
+---+
| a |
+---+
| 4 |
| 5 |
+---+
2 rows in set (0.00 sec)
|
将会话中的隔离界别改为RR，并删除a=4记录。
| 1 2 3 4 5 | MariaDB [db1]> set session transaction isolation level repeatable read;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> delete from db1.t1 where a=4;
Query OK, 1 row affected (0.00 sec)
|
| T1 RR | T2 RR |
|---|
| begin; | begin; |
| select * from db1.t1 where a>3 for update; | |
| 查询结果为5 | |
| | insert into db1.t1 values (4); |
| | ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction |
| | commit; |
| select * from db1.t1 where a>3; | |
| 查询结果为5 | |
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 | #事务T1
MariaDB [(none)]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [(none)]> select * from db1.t1 where a>3 for update;
+---+
| a |
+---+
| 5 |
+---+
1 row in set (0.02 sec)
#事务T2
MariaDB [(none)]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [(none)]> insert into db1.t1 values (4);
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
MariaDB [(none)]> commit;
Query OK, 0 rows affected (0.00 sec)
#事务T1
MariaDB [(none)]> select * from db1.t1 where a>3 for update;
+---+
| a |
+---+
| 5 |
+---+
1 row in set (0.02 sec)
|
认识锁的算法
nnoDB存储引擎的锁的算法有三种：
Lock的精度（type）分为 行锁、表锁、意向锁
Lock的模式（mode）分为:
知识点
-
innodb对于行的查询使用next-key lock
-
Next-locking keying为了解决Phantom Problem幻读问题
-
当查询的索引含有唯一属性时，将next-key lock降级为record key
-
Gap锁设计的目的是为了阻止多个事务将记录插入到同一范围内，而这会导致幻读问题的产生
-
有两种方式显式关闭gap锁：（除了外键约束和唯一性检查外，其余情况仅使用record lock） A. 将事务隔离级别设置为RC B. 将参数innodb_locks_unsafe_for_binlog设置为1
实践1： 验证next-key lock降级为record key
创建db1.t1表，有列a和b，分别为char(10)和int型，并且b为key，注意b列为索引列，但并不是主键，因此不是唯一的。
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 | MariaDB [db1]> create table db1.t1 (a char(10),b int,key (b));
Query OK, 0 rows affected (0.03 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',1),('superman',3),('leo',5);
Query OK, 3 rows affected (0.15 sec)
Records: 3 Duplicates: 0 Warnings: 0
MariaDB [db1]> select * from db1.t1;
+----------+------+
| a | b |
+----------+------+
| batman | 1 |
| superman | 3 |
| leo | 5 |
+----------+------+
3 rows in set (0.02 sec)
|
接下来开启两个事务T1和T2，T1中查看b=3的行，显式加排他锁；T1未提交事务时，T2事务开启并尝试插入新行a='batman',b=2和a='batman',b=4；
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 | #事务T1
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> select * from db1.t1 where b=3 for update;
+----------+------+
| a | b |
+----------+------+
| superman | 3 |
+----------+------+
1 row in set (0.12 sec)
#事务T2
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',2);
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
MariaDB [db1]> insert into db1.t1 values ('batman',4);
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
|
发现T2事务中不能插入新行a='batman',b=2和a='batman',b=4；可以查看当前innodb锁的信息
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 | MariaDB [db1]> select * from information_schema.innodb_locks\G;
*************************** 1. row ***************************
lock_id: 111B:0:334:3
lock_trx_id: 111B
lock_mode: X,GAP
lock_type: RECORD
lock_table: `db1`.`t1`
lock_index: `b`
lock_space: 0
lock_page: 334
lock_rec: 3
lock_data: 3, 0x00000000020E
*************************** 2. row ***************************
lock_id: 111A:0:334:3
lock_trx_id: 111A
lock_mode: X
lock_type: RECORD
lock_table: `db1`.`t1`
lock_index: `b`
lock_space: 0
lock_page: 334
lock_rec: 3
lock_data: 3, 0x00000000020E
2 rows in set (0.01 sec)
ERROR: No query specified
MariaDB [db1]> select * from information_schema.innodb_lock_waits\G;
*************************** 1. row ***************************
requesting_trx_id: 111B
requested_lock_id: 111B:0:334:3
blocking_trx_id: 111A
blocking_lock_id: 111A:0:334:3
1 row in set (0.09 sec)
MariaDB [db1]> select * from information_schema.innodb_lock_waits\G;
*************************** 1. row ***************************
requesting_trx_id: 111B
requested_lock_id: 111B:0:334:4
blocking_trx_id: 111A
blocking_lock_id: 111A:0:334:4
1 row in set (0.00 sec)
ERROR: No query specified
MariaDB [db1]> select * from information_schema.innodb_locks\G;
*************************** 1. row ***************************
lock_id: 111B:0:334:4
lock_trx_id: 111B
lock_mode: X,GAP
lock_type: RECORD
lock_table: `db1`.`t1`
lock_index: `b`
lock_space: 0
lock_page: 334
lock_rec: 4
lock_data: 5, 0x00000000020F
*************************** 2. row ***************************
lock_id: 111A:0:334:4
lock_trx_id: 111A
lock_mode: X,GAP
lock_type: RECORD
lock_table: `db1`.`t1`
lock_index: `b`
lock_space: 0
lock_page: 334
lock_rec: 4
lock_data: 5, 0x00000000020F
2 rows in set (0.11 sec)
ERROR: No query specified
|
我们看到T2事务的两次插入动作都在请求排他锁，但是此时T1事务已经在加了next-key lock(record + gap)，表现范围为b的(1,5)，包括记录3，所以T2事务在T1事务解锁之间，不能插入到b的(1,5)范围内
* lock_mode: X,GAP lock_mode 可以理解为 读锁还是写锁？；是在什么范围上锁？;此处加的写锁即排他锁；范围是(1,5)
* lock_type: RECORD 表示锁的精度，根据存储引擎不同，innodb是行锁，MYISAM是表锁
删除db1.t1表，重新创建db1.t1表，有列a和b，分别为char(10)和int型，并且b为primay key，因此b列是唯一的。
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 | MariaDB [db1]> drop tables t1;
Query OK, 0 rows affected (0.12 sec)
MariaDB [db1]> create table db1.t1 (a char(10),b int ,primary key (b));
Query OK, 0 rows affected (0.02 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',1),('superman',3),('leo',5);
Query OK, 3 rows affected (0.12 sec)
Records: 3 Duplicates: 0 Warnings: 0
MariaDB [db1]> select * from db1.t1;
+----------+---+
| a | b |
+----------+---+
| batman | 1 |
| superman | 3 |
| leo | 5 |
+----------+---+
3 rows in set (0.08 sec)
|
接下来开启两个事务T1和T2，T1中查看b=3的行，显式加排他锁；T1未提交事务时，T2事务开启并尝试插入新行a='batman',b=2和a='batman',b=4；
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 | #事务T1
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> select * from db1.t1 where b=3 for update;
+----------+---+
| a | b |
+----------+---+
| superman | 3 |
+----------+---+
1 row in set (0.14 sec)
#事务T2
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',2);
Query OK, 1 row affected (0.00 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',4);
Query OK, 1 row affected (0.00 sec)
|
继续在T2事务中尝试查看b=3的行，显式加共享锁。
| 1 2 3 | #事务T2
MariaDB [db1]> select * from db1.t1 where b=3 lock in share mode;
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
|
发现T2事务中可以插入新行a='batman',b=2和a='batman',b=4；但是不能查看b=3的行，接下来我们查看当前innodb锁的信息
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 | MariaDB [db1]> select * from information_schema.innodb_locks\G;
*************************** 1. row ***************************
lock_id: 1122:0:337:3
lock_trx_id: 1122
lock_mode: S
lock_type: RECORD
lock_table: `db1`.`t1`
lock_index: `PRIMARY`
lock_space: 0
lock_page: 337
lock_rec: 3
lock_data: 3
*************************** 2. row ***************************
lock_id: 1121:0:337:3
lock_trx_id: 1121
lock_mode: X
lock_type: RECORD
lock_table: `db1`.`t1`
lock_index: `PRIMARY`
lock_space: 0
lock_page: 337
lock_rec: 3
lock_data: 3
2 rows in set (0.02 sec)
ERROR: No query specified
MariaDB [db1]> select * from information_schema.innodb_lock_waits\G;
*************************** 1. row ***************************
requesting_trx_id: 1122
requested_lock_id: 1122:0:337:3
blocking_trx_id: 1121
blocking_lock_id: 1121:0:337:3
1 row in set (0.00 sec)
ERROR: No query specified
|
从以上信息可以看到，T1事务当前只在b=3所在的行上加了写锁，排他锁，并没有同时使用gap锁来组成next-key lock。
到此，已经证明了，当查询的索引含有唯一属性时，将next-key lock降级为record key
我们第二次创建的t1表的列b是主键，而主键必须是唯一的。
实践2： 关闭GAP锁_RC
有两种方式显式关闭gap锁：（除了外键约束和唯一性检查外，其余情况仅使用record lock）
A. 将事务隔离级别设置为RC B. 将参数innodb_locks_unsafe_for_binlog设置为1
| T1 RR | T2 RR |
|---|
| begin; | begin; |
| select * from db1.t1 where b=3 for update; | |
| | insert into db1.t1 values ('batman',2) |
| | ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction |
| | set session transaction isolation level READ COMMITTED; |
| commit; | commit; |
注意，将T1事务设置为RC后，需要将二进制日志的格式改为row格式，否则执行显式加锁时会报错
| 1 2 | MariaDB [db1]> insert into t1 values ('batman',2);
ERROR 1665 (HY000): Cannot execute statement: impossible to write to binary log since BINLOG_FORMAT = STATEMENT and at least one table uses a storage engine limited to row-based logging. InnoDB is limited to row-logging when transaction isolation level is READ COMMITTED or READ UNCOMMITTED.
|
| T1 RC | T2 RR |
|---|
| begin; | begin; |
| set session transaction isolation level READ COMMITTED; | |
| select * from db1.t1 where b=3 for update; | |
| | insert into db1.t1 values ('batman',2) |
| | insert into db1.t1 values ('batman',4) |
| commit; | commit; |
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 | #T1事务
MariaDB [db1]> set session transaction isolation level READ COMMITTED;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> select @@tx_isolation;
+----------------+
| @@tx_isolation |
+----------------+
| READ-COMMITTED |
+----------------+
1 row in set (0.00 sec)
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.09 sec)
MariaDB [db1]> select * from t1 where b=3 for update;
+----------+------+
| a | b |
+----------+------+
| superman | 3 |
+----------+------+
1 row in set (0.00 sec)
#T2事务
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.16 sec)
MariaDB [db1]> select @@tx_isolation;
+----------------+
| @@tx_isolation |
+----------------+
| READ-COMMITTED |
+----------------+
1 row in set (0.00 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',2);
Query OK, 1 row affected (0.00 sec)
MariaDB [db1]> commit;
Query OK, 0 rows affected (0.01 sec)
MariaDB [db1]> set session transaction isolation level REPEATABLE READ;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> select @@tx_isolation;
+-----------------+
| @@tx_isolation |
+-----------------+
| REPEATABLE-READ |
+-----------------+
1 row in set (0.00 sec)
MariaDB [db1]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',4);
Query OK, 1 row affected (0.00 sec)
MariaDB [db1]> commit;
Query OK, 0 rows affected (0.00 sec)
#T1事务
MariaDB [db1]> commit;
Query OK, 0 rows affected (0.00 sec)
|
我在做测试的时候，T1事务隔离界别为RC，T2事务的隔离界别分别用RC和RR做了测试，都是可以的
实践3： 关闭GAP锁_innodb_locks_unsafe_for_binlog
查看当前innodb_locks_unsafe_for_binlog参数的值
| 1 2 3 4 5 6 7 | MariaDB [(none)]> select @@innodb_locks_unsafe_for_binlog;
+----------------------------------+
| @@innodb_locks_unsafe_for_binlog |
+----------------------------------+
| 0 |
+----------------------------------+
1 row in set (0.00 sec)
|
修改参数，并重新启动服务
| 1 2 3 4 5 6 7 8 9 10 | [root@localhost ~]# vim /etc/my.cnf
innodb_locks_unsafe_for_binlog=1
[root@localhost ~]# systemctl restart mariadb
[root@localhost ~]# mysql -e "select @@innodb_locks_unsafe_for_binlog"
+----------------------------------+
| @@innodb_locks_unsafe_for_binlog |
+----------------------------------+
| 1 |
+----------------------------------+
|
还是去创建db1.t1表，如果已有就先drop；有列a和b，分别为char(10)和int型，并且b为key，注意b列为索引列，但并不是主键，因此不是唯一的。
| T1 RR | T2 RR |
|---|
| begin; | begin; |
| select * from db1.t1 where b=3 for update; | |
| | insert into db1.t1 values ('batman',2) |
| | insert into db1.t1 values ('batman',4) |
| commit; | commit; |
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 | MariaDB [db1]> create table db1.t1 (a char(10),b int,key (b));
Query OK, 0 rows affected (0.03 sec)
MariaDB [db1]> insert into db1.t1 values ('batman',1),('superman',3),('leo',5);
Query OK, 3 rows affected (0.15 sec)
Records: 3 Duplicates: 0 Warnings: 0
MariaDB [db1]> select * from db1.t1;
+----------+------+
| a | b |
+----------+------+
| batman | 1 |
| superman | 3 |
| leo | 5 |
+----------+------+
3 rows in set (0.02 sec)
|
接下来开启两个事务T1和T2，T1中查看b=3的行，显式加排他锁；T1未提交事务时，T2事务开启并尝试插入新行a='batman',b=2和a='batman',b=4；
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 | #T1事务
MariaDB [(none)]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [(none)]> select * from db1.t1 where b=3 for update;
+----------+------+
| a | b |
+----------+------+
| superman | 3 |
+----------+------+
1 row in set (0.01 sec)
#T2事务
MariaDB [(none)]> begin;
Query OK, 0 rows affected (0.00 sec)
MariaDB [(none)]> insert into db1.t1 values ('batman',4);
Query OK, 1 row affected (0.01 sec)
MariaDB [(none)]> insert into db1.t1 values ('batman',2);
Query OK, 1 row affected (0.00 sec)
MariaDB [(none)]> commit;
Query OK, 0 rows affected (0.00 sec)
#T1事务
MariaDB [(none)]> commit;
Query OK, 0 rows affected (0.00 sec)
|
不转行的女工程师 https://booboowei.github.io/
InnoDB锁详解
扫一扫
979
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
