Database Learning (Four) - Transactions

I. Introduction to transactions

1. Normal commit, automatic submission

commit;

2.rollback

delete from xxx where xx;
delete from xxx where xx;
savepoint sp1;
delete from xxx where xx;
rollback to sp1;

3. ACID properties of the transaction

A：原子性，操作集合不可分割
C：一致性，经过N个操作，数据状态不会改变
I：隔离性，隔离性会导致效率降低，为了提高程序效率，可以设置隔离级别
隔离级别：读未提交，读已提交，可重复度，序列化
数据不一致问题：脏读、不可重复读、幻读
D：持久性，所有数据修改持久化到介质中，不会因为程序关闭，导致丢失

4. characteristics of a transaction, which is the most critical?

All features are designed to ensure consistency, consistency is the ultimate pursuit.

Consistency is achieved by atom, isolation, durability guaranteed.

The locking mechanism:

In order to address concurrent access, data inconsistencies, have to consider locking granularity,

Lock granularity, the higher the efficiency, the larger the particle size, the lower the efficiency, mostly row-level locks

Second, dirty reads, phantom read, non-repeatable read test

1. Open a command line

select @@autocommit
set autocommit=0

2. Data Preparation

create database tran;
use tran;
create table psn(id int primary key,name varchar(10)) engine=innodb;
insert into psn values(1,"zhangsan");
insert into psn values(2,"lisi");
insert into psn values(3,"wangwu");
commit;

3. Testing Services

--事务包含四个隔离级别：（从上往下隔离级别越来越高
read uncommitted; --读未提交
read commited; --读已提交
repeatable read; --可重复读
(seriable) --序列化执行,串行

4. Test 1: generating an analog read dirty read uncommitted

set session transaction isolation level read uncommitted;
A:start transaction;
B:start transaction;
A:select * from psn;
B:select * from psn;

A:update psn set name='msb'
--读取结果为msb,产生脏读，因为A事务没有commit

A:commit
B:select * from psn;
--此时读取的数据是msb,正常读取

At this time, we have two tables for comparison and found inconsistent data, generate dirty read phenomenon

5. Test 2: non-repeatable read analog

set session transaction isolation level read committed;
A:start transaction;
B:start transaction;
A:update psn set name='ppp'
B:select * from psn;
A:commit;
B:select * from psn;

At this time carried out by AB inquiry found no dirty reads, but the emergence of non-repeatable read, because in a single transaction, B read out two different data

6. Test 3: simulated phantom read

set session transaction isolation level repeatable read;
A:start transaction;
B:start transaction;
A:update psn set name='lisi'
B:select * from psn;
A:commit;
B:select * from psn;

This is the climate has been submitted, but B hallucinations, or prior to the data, so called phantom reads.

7. Summary Table

Isolation Levels	abnormal situation
Uncommitted Read	Dirty reads, repeatable read, phantom read
Read Committed	Repeatable read, phantom read
Non-repeatable read	Magic Reading
Serialization

Database Learning (Four) - Transaction Processing (dirty reads, non-repeatable read, phantom read this all in a)