MySQL InnoDB加锁超时回滚机制（转）

add by zhj: 看来我对MySQL的理解还有待深入，水还是挺深的啊，MySQL给记录加锁时，可以通过innodb_lock_wait_timeout参数设置超时时间，

如果加锁等待超过这个时间，就会回滚，但回滚的话有两种方式：第一种：回滚当前加锁的这条语句；第二种：回滚整个事务。这两种方式是通过参数

innodb_rollback_on_timeout来控制的。如果是OFF，表示加锁超时回滚时，只回滚加锁超时的那条SQL语句；如果是ON，表示回滚整个事务。默认

是OFF。在《MySQL Admin Cookbook》一书中，作者强烈建议该参数设置为ON，但其实OFF也没关系，你可以在应用程序中捕获那个加锁超时，然

后应用程序去执行ROLLBACK，这样就可以保证原子性，只要你没执行COMMIT，就不会破坏原子性。Django应该就是这么做的。我个人拙见是：感觉

设置为ON，依靠数据库本身自动完成回滚更好一些。

原文：http://www.cnblogs.com/hustcat/archive/2012/11/18/2775487.html

1、innodb_rollback_on_timeout变量

下面是MySQL官方手册关开innodb_rollback_on_timeout变量的说明：

In MySQL 5.0.13 and up, InnoDB rolls back only the last statement on a transaction timeout by default. If --innodb_rollback_on_timeout is specified, a transaction timeout causes InnoDB to abort and roll back the entire transaction (the same behavior as before MySQL 5.0.13). This variable was added in MySQL 5.0.32.

该变量默认值为OFF，如果事务因为加锁超时，会回滚上一条语句执行的操作。如果设置ON，则整个事务都会回滚。

下面通过一个示例来验证上面这段话。

2、示例

(1) innodb_rollback_on_timeout为OFF

Session 1	Session 2
mysql> create table tt(c1 int primary key, c2 int)engine=innodb; Query OK, 0 rows affected (0.01 sec) mysql> insert into tt values(1, 1); Query OK, 1 row affected (0.00 sec) mysql> begin; Query OK, 0 rows affected (0.00 sec) mysql> select * from tt where c1=1 lock in share mode; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | +----+------+ 1 row in set (0.00 sec)
	mysql> begin; Query OK, 0 rows affected (0.00 sec)   mysql> insert into tt values(10,10); Query OK, 1 row affected (0.00 sec)   mysql> delete from tt where c1=1; ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction mysql> select * from tt; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | | 10 |   10 | +----+------+ 2 rows in set (0.00 sec)   mysql> rollback; Query OK, 0 rows affected (0.01 sec)   mysql> select * from tt; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | +----+------+ 1 row in set (0.00 sec)
mysql> select * from tt; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | +----+------+ 1 row in set (0.00 sec)
	mysql> begin; Query OK, 0 rows affected (0.00 sec)   mysql> insert into tt values(10,10); Query OK, 1 row affected (0.00 sec)   mysql> delete from tt where c1=1; ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction mysql> commit; Query OK, 0 rows affected (0.02 sec)   mysql> select * from tt; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | | 10 |   10 | +----+------+ 2 rows in set (0.00 sec)

session2因为加锁超时，事务回退到上一条语句。 

(2) innodb_rollback_on_timeout为ON

Session 1	Session 2
mysql> begin; Query OK, 0 rows affected (0.00 sec) mysql> select * from tt where c1=1 lock in share mode; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | +----+------+ 1 row in set (0.00 sec)
	mysql> begin; Query OK, 0 rows affected (0.00 sec)   mysql> insert into tt values(11,11); Query OK, 1 row affected (0.00 sec)   mysql> delete from tt where c1=1; ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction mysql> select * from tt; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | | 10 |   10 | +----+------+ 2 rows in set (0.00 sec) mysql> commit; Query OK, 0 rows affected (0.00 sec)   mysql> select * from tt; +----+------+ | c1 | c2   | +----+------+ |  1 |    1 | | 10 |   10 | +----+------+ 2 rows in set (0.00 sec)

session2加锁超时，整个事务回滚。

3、总结

innodb_rollback_on_timeout为OFF，事务会回滚到上一个保存点，InnoDB在执行每条SQL语句之前，都会创建一个保存点，参见代码：

int

row_insert_for_mysql(

                                               /* out: error code or DB_SUCCESS */

         byte*                 mysql_rec,       /* in: row in the MySQL format */

         row_prebuilt_t*     prebuilt)  /* in: prebuilt struct in MySQL

                                               handle */

{

。。。

         savept = trx_savept_take(trx);

。。。

如果事务因为加锁超时，相当于回滚到上一条语句。但是报错后，事务还没有完成，用户可以选择是继续提交，或者回滚之前的操作，由用户选择是否进一步提交或者回滚事务。

innodb_rollback_on_timeout为ON，整个事务都会回滚。这可以从row_mysql_handle_errors函数中得到验证。

ibool
row_mysql_handle_errors(
/*====================*/
                /* out: TRUE if it was a lock wait and
                we should continue running the query thread */
    ulint*        new_err,/* out: possible new error encountered in
                lock wait, or if no new error, the value
                of trx->error_state at the entry of this
                function */
    trx_t*        trx,    /* in: transaction */
    que_thr_t*    thr,    /* in: query thread */
    trx_savept_t*    savept)    /* in: savepoint or NULL */
{
...
else if (err == DB_DEADLOCK //发生死锁
           || err == DB_LOCK_TABLE_FULL
           || (err == DB_LOCK_WAIT_TIMEOUT
               && row_rollback_on_timeout)) {
        /* Roll back the whole transaction; this resolution was added
        to version 3.23.43 */

        trx_general_rollback_for_mysql(trx, FALSE, NULL); //事务全部回滚
                
    } else if (err == DB_OUT_OF_FILE_SPACE
           || err == DB_LOCK_WAIT_TIMEOUT) {

        ut_ad(!(err == DB_LOCK_WAIT_TIMEOUT
                && row_rollback_on_timeout));

               if (savept) { //回滚到上一个保存点
            /* Roll back the latest, possibly incomplete
            insertion or update */

            trx_general_rollback_for_mysql(trx, TRUE, savept);
        }
        /* MySQL will roll back the latest SQL statement */
...

问题：innodb_rollback_on_timeout为OFF，事务的原子性被破坏了吗？

答：NO，从示例中可以看到，事务只是回退上一条语句的状态，而整个事务实际上没有完成(提交或者回滚)，而作为应用程序在检测这个错误时，应该选择是提交或者回滚事务。如果严格要求事务的原子性，当然是执行ROLLBACK，回滚事务。