数据库事务隔离级别：
| Isolation\Problem |dirty reads|non-repeatable reads|phantom reads |
|--------------------------------------------------------------------|
| READ_UNCOMMITTED | yes | yes | yes |
| READ_COMMITTED | no | yes | yes |
| REPEATABLE_READ | no | no | yes |
| SERIALIZABLE | no | no | no |
本文将以Spring + Hibernate + postgres 为例，重现各个级别遇到的问题。
主要代码：
Entity:
@Entity
public class City implements Serializable {
private static final long serialVersionUID = 1L;
@Id
@GeneratedValue(strategy = GenerationType.IDENTITY)
private Long id;
@Column(nullable = false)
private String name;
@Column(nullable = false)
private String country;
@OneToMany(mappedBy="city")
private Set hotles;
...
Service:
@Component("cityService")
@Transactional(isolation=Isolation.READ_UNCOMMITTED)
class SampleServiceImpl implements SampleService {
@Autowired
private SessionFactory sessionFactory;
@Override
public void request_1() {
Session session = sessionFactory.getCurrentSession();
City city = (City) session.get(City.class, 1L);
System.out.println(city);
session.clear();
city = (City) session.get(City.class, 1L);
System.out.println(city);
}
@Override
public void request_2() {
Session session = sessionFactory.getCurrentSession();
City city = (City) session.get(City.class, 1L);
city.setName("ShangHai");
city.setCountry("China");
session.flush();
System.out.println(city);
}
@Override
public void request_3() {
Session session = sessionFactory.getCurrentSession();
int totalCity = session.createCriteria(City.class).list().size();
System.out.println("First time: " + totalCity);
int totalCitySecondTime = session.createCriteria(City.class).list().size();
System.out.println("Second time: " + totalCitySecondTime);
}
@Override
public void request_4() {
Session session = sessionFactory.getCurrentSession();
City city = new City("BeiJing", "China");
session.save(city);
}
}
READ_UNCOMMITTED :
READ_UNCOMMITTED 是最低的事务级别，会出现"脏读"，"不可重复读"，"幻读"。
如上代码所示，使用@Transactional(isolation=Isolation.READ_UNCOMMITTED)已将事务的隔离级别设置成了READ_UNCOMMITTED。接下来，我们将在这个事务级别重现“脏读”。
重现步骤：
- 以Debug Mode启动程序，并在
request_1()中session.clear()处打上断点。 - 触发
request_1()，让当前线程停留在session.clear()处。 - 在
request_2()中System.out.println(city)处打上断点，然后触发request_2()，让其更新City数据，但并不提交事务（停在断点处，不让方法返回） - 让
request_1()线程从session.clear()处继续执行。
期望结果：
request_1()两次打印出的City数据应该不一样。（在此级别下出现了"脏读"，request_1()读取到了request_2()未提交事务的修改。）
实际结果：
2017-02-26 11:36:23.274 DEBUG 6948 --- [nio-8080-exec-1] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
City [name=Brisbane, country=Australia]
2017-02-26 11:36:28.079 DEBUG 6948 --- [nio-8080-exec-2] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
2017-02-26 11:36:28.108 DEBUG 6948 --- [nio-8080-exec-2] org.hibernate.SQL : update City set country=?, name=? where id=?
Hibernate: update City set country=?, name=? where id=?
2017-02-26 11:36:45.977 DEBUG 6948 --- [nio-8080-exec-1] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
City [name=Brisbane, country=Australia]
[nio-8080-exec-1]是request_1()打印的结果，[nio-8080-exec-2]是request_2()打印的结果。
request_1()两次都打印出了City [name=Brisbane, country=Australia]，与期望不同，并没出现"脏读"。
找到了一篇文章对该现象做出了解释:
虽然官方宣称PostgreSQL支持所有四种ANSI事务隔离级别，但事实上PostgreSQL中只有三种事务隔离级别。每当查询请求“未提交读”时，PostgreSQL就默默地将其升级为“提交读”。因此PostgreSQL不允许脏读。
结论：
使用Postgres时，使用@Transactional(isolation=Isolation.READ_UNCOMMITTED)将不会生效，事务隔离级别将自动提升为@Transactional(isolation=Isolation.READ_COMMITTED)，所以"脏读"也是不会出现的.
READ_COMMITTED:
READ_COMMITTED解决了“脏读“问题，在一个事务中是不会读取到其它事务未提交的修改的。也就是上一个例子中request_1()两次打印结果都相同，并不会读取到request_2()未提交的修改。
但READ_COMMITTED会出现“不可重复读”问题。接下来，我们将在这个事务级别重现“不可重复读”
重现步骤：
- 改为
@Transactional(isolation=Isolation.READ_COMMITTED) - 以Debug Mode重新启动程序(让spring重新导入数据)，并在
request_1()中session.clear()处打上断点。 - 触发
request_1()，让当前线程停留在session.clear()处。 - 触发
request_2()，让其更新City数据，并让其返回提交事务。 - 让
request_1()线程从session.clear()处继续执行。
期望结果：
request_1()两次打印出的City数据应该不一样。（在此级别下出现了"不可重复读"，request_1()两次访问数据库拿到的数据不一致）
实际结果：
2017-02-26 12:07:59.964 DEBUG 6912 --- [nio-8080-exec-1] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
City [name=Brisbane, country=Australia]
2017-02-26 12:08:06.207 DEBUG 6912 --- [nio-8080-exec-2] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
2017-02-26 12:08:06.241 DEBUG 6912 --- [nio-8080-exec-2] org.hibernate.SQL : update City set country=?, name=? where id=?
Hibernate: update City set country=?, name=? where id=?
City [name=ShangHai, country=China]
2017-02-26 12:08:16.576 DEBUG 6912 --- [nio-8080-exec-1] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
City [name=ShangHai, country=China]
[nio-8080-exec-1]是request_1()打印的结果，[nio-8080-exec-2]是request_2()打印的结果。
request_1()两次打印结果的确不同，表明在request_1()中出现了“不可重复读”
结论：
@Transactional(isolation=Isolation.READ_COMMITTED)可能会出现同个事务中两次拿同条数据，而两次拿到的数据不一致问题，这即是“不可重复读”。为了避免“不可重复读”，我们可以将事务隔离级别提高到REPEATABLE_READ(改为@Transactional(isolation=Isolation.REPEATABLE_READ))，然后再按照上面的步骤进行试验，得到结果：
2017-02-26 12:14:33.065 DEBUG 5372 --- [nio-8080-exec-1] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
City [name=Brisbane, country=Australia]
2017-02-26 12:14:37.424 DEBUG 5372 --- [nio-8080-exec-2] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
2017-02-26 12:14:37.456 DEBUG 5372 --- [nio-8080-exec-2] org.hibernate.SQL : update City set country=?, name=? where id=?
Hibernate: update City set country=?, name=? where id=?
City [name=ShangHai, country=China]
2017-02-26 12:14:43.559 DEBUG 5372 --- [nio-8080-exec-1] org.hibernate.SQL : select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
Hibernate: select city0_.id as id1_0_0_, city0_.country as country2_0_0_, city0_.name as name3_0_0_ from City city0_ where city0_.id=?
City [name=Brisbane, country=Australia]
结果中可以看到request_1()两次都打印出了City [name=Brisbane, country=Australia]，两次读取数据一致了，避免了“不可重复读”。
REPEATABLE_READ:
REPEATABLE_READ解决了“不可重复读“问题。
但REPEATABLE_READ会出现“幻读”问题。接下来，我们将在这个事务级别重现““幻读””
重现步骤：
期望结果：
实际结果：
Code:
Sample Code on Github
参考：
How Does Spring @Transactional Really Work?
Spring transaction isolation level tutorial
理解事务的4种隔离级别
Spring @Transactional(isolation=Isolation.REPEATABLE_READ) not work