ITPub博客

首页 > 数据库 > MySQL > mysqldump --single-transaction一致性的研究

mysqldump --single-transaction一致性的研究

原创 MySQL 作者:水逸冰 时间:2019-08-02 19:17:41 0 删除 编辑

--single-transaction 的含义是,在开始 dump 前,设置隔离级别为 RR ,并且 start transaction. 通过将导出操作封装在一个事务 (Repeatable Read) 内来使得导出的数据是一个一致性快照。

       测试数据库版本:

mysql> select version();

+------------+

| version()  |

+------------+

| 5.7.26-log |

+------------+

1 row in set (0.00 sec)

测试 1  

sesseion A

session B

mysql> set tx_isolation='repeatable-read';

Query OK, 0 rows affected (0.00 sec)

 

mysql> set tx_isolation='repeatable-read';

Query OK, 0 rows affected (0.00 sec)

 

mysql> begin;

Query OK, 0 rows affected (0.01 sec)



mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

+----+------+

3 rows in set (0.00 sec) 

 

mysql> insert into test01 select 4,4;

Query OK, 1 row affected (0.00 sec)

Records: 1  Duplicates: 0  Warnings: 0

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

|    4 |    4 |

+----+------+

4 rows in set (0.00 sec)



mysql> insert into test01 select 5,5;

Query OK, 1 row affected (0.00 sec)

Records: 1  Duplicates: 0  Warnings: 0

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

|    4 |    4 |

+----+------+

4 rows in set (0.00 sec)


       可以看到, RR 隔离级别下的一致性读,不是以事务 begin 的时间点建立 snapshot 的。单纯 begin 后,到第一次读取前之间其他会话的事务还是可以读取的,但是在第一次读取数据完成后,读取到的事务就不再变化了。

测试 2

session A

session B

mysql> set tx_isolation='repeatable-read';

mysql> set tx_isolation='repeatable-read';


mysql> select * from t1;

Empty set (0.00 sec)

mysql> begin;

Query OK, 0 rows affected (0.00 sec)

 

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

|    4 |    4 |

|    5 |    5 |

+----+------+

5 rows in set (0.00 sec)



mysql> insert into test01 select 6,6;

Query OK, 1 row affected (0.00 sec)

Records: 1  Duplicates: 0  Warnings: 0

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

|    4 |    4 |

|    5 |    5 |

+----+------+

5 rows in set (0.00 sec)


       RR 隔离级别下的一致性读,发生在事务里面第一次 select 的时候。我试了下用 delete 语句替换 session A select 语句,结果发现还是会读取到 session B 的事务。

session A

session B

mysql> set tx_isolation='repeatable-read';

mysql> set tx_isolation='repeatable-read';

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

+----+------+

3 rows in set (0.00 sec)

mysql> begin;

Query OK, 0 rows affected (0.00 sec)

 

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

+----+------+

3 rows in set (0.00 sec)



mysql> insert into test01 select 4,4;

Query OK, 1 row affected (0.00 sec)

Records: 1  Duplicates: 0  Warnings: 0

 

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

|    4 |    4 |

+----+------+

4 rows in set (0.00 sec)

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

+----+------+

3 rows in set (0.00 sec)


mysql> update test01 set c2=5 where   c1=4;

Query OK, 1 row affected (0.00 sec)

Rows matched: 1  Changed: 1    Warnings: 0

 

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

|    4 |    5 |

+----+------+

4 rows in set (0.00 sec)


       Session A 第一次 select 建立一致性读后, session B 插入数据, session A select 仍然读不到,但是 update 因为是当前读,所以更新到 session B 插入的数据。

       所以 begin start transaction 是事务开始的标志,但不是事务开始的起点。如果要将 start transaction 作为事务开始的时间点,那么必须使用:

START TRANSACTION WITH consistent snapshot    ###mysqldump 中的快照就是用这个实现的

mysql> set tx_isolation='repeatable-read';

mysql> set tx_isolation='repeatable-read';


mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

+----+------+

3 rows in set (0.00 sec)

mysql> start transaction with   consistent snapshot;

Query OK, 0 rows affected (0.00 sec)



mysql> insert into test01 select 4,4;

Query OK, 1 row affected (0.00 sec)

Records: 1  Duplicates: 0  Warnings: 0

mysql> select * from test01;

+----+------+

| c1 | c2   |

+----+------+

|    1 |    1 |

|    2 |    2 |

|    3 |    3 |

+----+------+

3 rows in set (0.00 sec)


发起备份

mysqldump -uroot -poracle  --single-transaction --master-data=1 -R -E --triggers -B ming --ignore_table=ming.test02 > /tmp/ming_st.sql

查看备份过程的 general log

[root@oradb-2062 binlog]# more   /u01/mysql/3306/data/oradb-2062.log

/u01/mysql_57/bin/mysqld, Version:   5.7.26-log (MySQL Community Server (GPL)). started with:

Tcp port: 3306  Unix socket:   /u01/mysql/3306/data/mysqld.sock

Time                 Id Command    Argument

2019-08-01T08:30:50.718358Z        12 Query     show variables like 'log_output'

2019-08-01T08:31:33.211254Z        14 Connect   root@localhost on  using Socket

2019-08-01T08:31:33.211413Z        14 Query     /*!40100 SET @@SQL_MODE='' */

2019-08-01T08:31:33.211474Z        14 Query     /*!40103 SET TIME_ZONE='+00:00' */

2019-08-01T08:31:33.211565Z        14 Query     FLUSH /*!40101 LOCAL */ TABLES

2019-08-01T08:31:33.212009Z        14 Query     FLUSH TABLES WITH READ LOCK

2019-08-01T08:31:33.212047Z        14 Query     SET SESSION TRANSACTION ISOLATION LEVEL REPEATABLE READ

2019-08-01T08:31:33.212070Z        14 Query     START TRANSACTION /*!40100 WITH CONSISTENT SNAPSHOT */

2019-08-01T08:31:33.212115Z        14 Query     SHOW VARIABLES LIKE 'gtid\_mode'

2019-08-01T08:31:33.216296Z        14 Query     SHOW MASTER STATUS

2019-08-01T08:31:33.216472Z        14 Query     UNLOCK TABLES

2019-08-01T08:31:33.219582Z        14 Query     SELECT LOGFILE_GROUP_NAME, FILE_NAME,   TOTAL_EXTENTS, INITIAL_SIZE, ENGINE, EXTRA FROM INFORMATION_SCHEMA.FILES   WHERE FILE_TYPE = 'UNDO LOG

' AND FILE_NAME IS NOT NULL AND   LOGFILE_GROUP_NAME IS NOT NULL AND LOGFILE_GROUP_NAME IN (SELECT DISTINCT   LOGFILE_GROUP_NAME FROM INFORMATION_SCHEMA.FILES WHERE FILE_TYPE = 'DATAFILE'   AN

D TABLESPACE_NAME IN (SELECT DISTINCT   TABLESPACE_NAME FROM INFORMATION_SCHEMA.PARTITIONS WHERE TABLE_SCHEMA IN   ('ming'))) GROUP BY LOGFILE_GROUP_NAME, FILE_NAME, ENGINE, TOTAL_EXTENTS, I

NITIAL_SIZE ORDER BY LOGFILE_GROUP_NAME

2019-08-01T08:31:33.223184Z        14 Query     SELECT DISTINCT TABLESPACE_NAME,   FILE_NAME, LOGFILE_GROUP_NAME, EXTENT_SIZE, INITIAL_SIZE, ENGINE FROM   INFORMATION_SCHEMA.FILES WHERE FILE

_TYPE = 'DATAFILE' AND TABLESPACE_NAME IN   (SELECT DISTINCT TABLESPACE_NAME FROM INFORMATION_SCHEMA.PARTITIONS WHERE   TABLE_SCHEMA IN ('ming')) ORDER BY TABLESPACE_NAME, LOGFILE_GROUP_NAME

2019-08-01T08:31:33.223727Z        14 Query     SHOW VARIABLES LIKE 'ndbinfo\_version'

2019-08-01T08:31:33.225502Z        14 Init DB   ming

2019-08-01T08:31:33.225545Z        14 Query     SHOW CREATE DATABASE IF NOT EXISTS   `ming`

2019-08-01T08:31:33.225683Z        14 Query     SAVEPOINT sp

2019-08-01T08:31:33.225750Z        14 Query     show tables

2019-08-01T08:31:33.225957Z        14 Query     show table status like 'mytest01'

2019-08-01T08:31:33.226083Z        14 Query     SET SQL_QUOTE_SHOW_CREATE=1

2019-08-01T08:31:33.226114Z        14 Query     SET SESSION character_set_results =   'binary'

2019-08-01T08:31:33.226145Z        14 Query     show create table `mytest01`

2019-08-01T08:31:33.226190Z        14 Query     SET SESSION character_set_results =   'utf8'

2019-08-01T08:31:33.226226Z        14 Query     show fields from `mytest01`

2019-08-01T08:31:33.226468Z        14 Query     show fields from `mytest01`

2019-08-01T08:31:33.226687Z        14 Query     SELECT /*!40001 SQL_NO_CACHE */ * FROM   `mytest01`

2019-08-01T08:31:33.226810Z        14 Query     SET SESSION character_set_results =   'binary'

2019-08-01T08:31:33.226844Z        14 Query     use `ming`

2019-08-01T08:31:33.226877Z        14 Query     select @@collation_database

2019-08-01T08:31:33.226920Z        14 Query     SHOW TRIGGERS LIKE 'mytest01'

2019-08-01T08:31:33.227098Z        14 Query     SET SESSION character_set_results =   'utf8'

2019-08-01T08:31:33.227130Z        14 Query     ROLLBACK TO SAVEPOINT sp

2019-08-01T08:31:33.227162Z        14 Query     show table status like 'test01'

2019-08-01T08:31:33.227262Z        14 Query     SET SQL_QUOTE_SHOW_CREATE=1

2019-08-01T08:31:33.227289Z        14 Query     SET SESSION character_set_results =   'binary'

2019-08-01T08:31:33.227316Z        14 Query     show create table `test01`

2019-08-01T08:31:33.227356Z        14 Query     SET SESSION character_set_results =   'utf8'

2019-08-01T08:31:33.227389Z        14 Query     show fields from `test01`

2019-08-01T08:31:33.227730Z        14 Query     show fields from `test01`

2019-08-01T08:31:33.227911Z        14 Query     SELECT /*!40001 SQL_NO_CACHE */ * FROM   `test01`

2019-08-01T08:31:33.228005Z        14 Query     SET SESSION character_set_results =   'binary'

2019-08-01T08:31:33.228053Z        14 Query     use `ming`

2019-08-01T08:31:33.228084Z        14 Query     select @@collation_database

2019-08-01T08:31:33.228143Z        14 Query     SHOW TRIGGERS LIKE 'test01'

2019-08-01T08:31:33.228336Z        14 Query     SET SESSION character_set_results =   'utf8'

2019-08-01T08:31:33.228369Z        14 Query     ROLLBACK TO SAVEPOINT sp

2019-08-01T08:31:33.228399Z        14 Query     show table status like 'test03'

2019-08-01T08:31:33.228501Z        14 Query     SET SQL_QUOTE_SHOW_CREATE=1

2019-08-01T08:31:33.228686Z        14 Query     SET SESSION character_set_results =   'binary'

2019-08-01T08:31:33.228726Z        14 Query     show create table `test03`

。。。。。。

可以看到, mysqldump 的大致实现过程是:连接 -> 初始化信息 -> 刷新表(锁表) -> 开启事务(一致性快照) -> 记录偏移量 -> 解锁表


参考:https://yq.aliyun.com/articles/552972?spm=a2c4e.11153940.0.0.18a12415csEBjM

来自 “ ITPUB博客 ” ,链接:http://blog.itpub.net/31480688/viewspace-2652628/,如需转载,请注明出处,否则将追究法律责任。

上一篇: mysql锁之死锁
请登录后发表评论 登录
全部评论
精通oracle和linux,热衷于研究oracle,擅长shell和Python编程,喜欢自动化运维。

注册时间:2017-08-05

  • 博文量
    91
  • 访问量
    77705