【飞天奔月出品】聊聊JPA之GenerationType.AUTO

飞天奔月

浏览: 520221 次
性别:
来自: 上海

最近访客更多访客>>

chao_t

xinxiangli

Comven_eye

天意之都

博主相关

博客

微博

相册

留言

关于我

文章分类

社区版块

存档分类

博客分类：

JPA

多说 GenerationType.AUTO 适用于多个数据库,

写道

在我们的应用中，一般选用@GeneratedValue(strategy=GenerationType.AUTO)这种方式，自动选择主键生成策略，以适应不同的数据库移植。

为什么我今天玩 sqlserver 可以创建主键, 但是主键却没有 IDENTITY标识呢?

难道是老人说的是错的? 难道教科书上写得不对? 事出反常必有妖,我偏偏不信这个邪

1.现象

现在这么一个简单的JPA类

@Entity
@Table(name = "T_SYS_CHOOSE_OPTION")
public class ChooseOption extends BaseModel implements Command{

    private Long id;

    @Id
    @Column(name = "ID")
    @GeneratedValue(strategy = GenerationType.AUTO)
    public Long getId(){
        return id;
    }

    public void setId(Long id){
        this.id = id;
    }
}

但是启动程序, 生成的表,id列却没有 IDENTITY 标识

2.调试

将 logback 相关类日志调整成 debug

<logger name="org.hibernate.tool.hbm2ddl.SchemaUpdate" level="DEBUG" />

对比下 GenerationType.AUTO 与 GenerationType.IDENTITY 生成的SQL语句

--GenerationType.IDENTITY

create table T_SYS_CHOOSE_OPTION (ID numeric(19,0) identity not null)

--GenerationType.AUTO

create table T_SYS_CHOOSE_OPTION (ID numeric(19,0) not null)

很清晰的发现,生成的SQL不同点

3.分析问题

上面的想象,搜索了 google ,stackoverflow 等网站, 有用的文章不是很多

那我只有自己动手了, 拿出绝招, 那就是源代码跟踪

最终定位到 org.hibernate.mapping.Table.sqlCreateString(Dialect, Mapping, String, String) 方法

核心代码片段:

public String sqlCreateString(Dialect dialect, Mapping p, String defaultCatalog, String defaultSchema) {
		StringBuffer buf = new StringBuffer( hasPrimaryKey() ? dialect.getCreateTableString() : dialect.getCreateMultisetTableString() )
				.append( ' ' )
				.append( getQualifiedName( dialect, defaultCatalog, defaultSchema ) )
				.append( " (" );

		boolean identityColumn = idValue != null && idValue.isIdentityColumn( p.getIdentifierGeneratorFactory(), dialect );

		// Try to find out the name of the primary key to create it as identity if the IdentityGenerator is used
		String pkname = null;
		if ( hasPrimaryKey() && identityColumn ) {
			pkname = ( (Column) getPrimaryKey().getColumnIterator().next() ).getQuotedName( dialect );
		}

		Iterator iter = getColumnIterator();
		while ( iter.hasNext() ) {
			Column col = (Column) iter.next();

			buf.append( col.getQuotedName( dialect ) )
					.append( ' ' );

			if ( identityColumn && col.getQuotedName( dialect ).equals( pkname ) ) {
				// to support dialects that have their own identity data type
				if ( dialect.hasDataTypeInIdentityColumn() ) {
					buf.append( col.getSqlType( dialect, p ) );
				}
				buf.append( ' ' )
						.append( dialect.getIdentityColumnString( col.getSqlTypeCode( p ) ) );
			}
			else {

.....

生成 Identity 的关键语句是

buf.append( ' ' )
						.append( dialect.getIdentityColumnString( col.getSqlTypeCode( p ) ) );

我这里 dialect 配置的是

hibernate.dialect=org.hibernate.dialect.SQLServerDialect

该方言 getIdentityColumnString 方法的实现如下:

可以看出 GenerationType.IDENTITY 的实现原理是拼接了 identity not null

那为什么 GenerationType.AUTO 就没有拼接呢? 难道是我的人品不好?

别急,还有个关键代码是上面的 if 流程判断

if ( identityColumn && col.getQuotedName( dialect ).equals( pkname ) ) {

在于

boolean identityColumn = idValue != null && idValue.isIdentityColumn( p.getIdentifierGeneratorFactory(), dialect );

而我当 GenerationType.AUTO , 调试到这个地方的时候, identityColumn =false

进一步跟踪发现

org.hibernate.mapping.SimpleValue.isIdentityColumn(IdentifierGeneratorFactory, Dialect)

	public boolean isIdentityColumn(IdentifierGeneratorFactory identifierGeneratorFactory, Dialect dialect) {
		identifierGeneratorFactory.setDialect( dialect );
		return identifierGeneratorFactory.getIdentifierGeneratorClass( identifierGeneratorStrategy )
				.equals( IdentityGenerator.class );
	}

其中的 identifierGeneratorStrategy 值为 org.hibernate.id.enhanced.SequenceStyleGenerator

导致下面的代码,并没有使用 native 去取到 dialect.getNativeIdentifierGeneratorClass()

	/**
	 * {@inheritDoc}
	 */
	public Class getIdentifierGeneratorClass(String strategy) {
		if ( "native".equals( strategy ) ) {
			return dialect.getNativeIdentifierGeneratorClass();
		}

		Class generatorClass = ( Class ) generatorStrategyToClassNameMap.get( strategy );
		try {
			if ( generatorClass == null ) {
				generatorClass = ReflectHelper.classForName( strategy );
			}
		}
		catch ( ClassNotFoundException e ) {
			throw new MappingException( "Could not interpret id generator strategy [" + strategy + "]" );
		}
		return generatorClass;
	}

4. 为毛 identifierGeneratorStrategy 值为 org.hibernate.id.enhanced.SequenceStyleGenerator

追本溯源 , 我们来到了 generatorType 定义的地方

org.hibernate.cfg.AnnotationBinder.processId(PropertyHolder, PropertyData, SimpleValue, HashMap<String, IdGenerator>, boolean, ExtendedMappings)

	private static void processId(
			PropertyHolder propertyHolder,
			PropertyData inferredData,
			SimpleValue idValue,
			HashMap<String, IdGenerator> classGenerators,
			boolean isIdentifierMapper,
			ExtendedMappings mappings) {
.........


		String generatorType = generatedValue != null ?
				generatorType( generatedValue.strategy(), mappings ) :
				"assigned";
		String generatorName = generatedValue != null ?
				generatedValue.generator() :
				BinderHelper.ANNOTATION_STRING_DEFAULT;
		if ( isComponent ) {
			generatorType = "assigned";
		} //a component must not have any generator
		BinderHelper.makeIdGenerator( idValue, generatorType, generatorName, mappings, localGenerators );

.....

以及 org.hibernate.cfg.AnnotationBinder.generatorType(GenerationType, ExtendedMappings)

	private static String generatorType(GenerationType generatorEnum, ExtendedMappings mappings) {
		boolean useNewGeneratorMappings = mappings.useNewGeneratorMappings();
		switch ( generatorEnum ) {
			case IDENTITY:
				return "identity";
			case AUTO:
				return useNewGeneratorMappings
						? org.hibernate.id.enhanced.SequenceStyleGenerator.class.getName()
						: "native";
			case TABLE:
				return useNewGeneratorMappings
						? org.hibernate.id.enhanced.TableGenerator.class.getName()
						: MultipleHiLoPerTableGenerator.class.getName();
			case SEQUENCE:
				return useNewGeneratorMappings
						? org.hibernate.id.enhanced.SequenceStyleGenerator.class.getName()
						: "seqhilo";
		}
		throw new AssertionFailure( "Unknown GeneratorType: " + generatorEnum );
	}

哦, 原来在这里

			case AUTO:
				return useNewGeneratorMappings
						? org.hibernate.id.enhanced.SequenceStyleGenerator.class.getName()
						: "native";

发现, 当 AUTO 的时候, 如果 useNewGeneratorMappings为true的话 ,那么就会使用 SequenceStyleGenerator 而不是传说中的 native

那么这里的 useNewGeneratorMappings 是否可以变更呢?

看下 useNewGeneratorMappings 的源码 org.hibernate.cfg.AnnotationConfiguration.ExtendedMappingsImpl.useNewGeneratorMappings()

		public boolean useNewGeneratorMappings() {
			if ( useNewGeneratorMappings == null ) {
				final String booleanName = getConfigurationProperties().getProperty( USE_NEW_ID_GENERATOR_MAPPINGS );
				useNewGeneratorMappings = Boolean.valueOf( booleanName );
			}
			return useNewGeneratorMappings.booleanValue();
		}

hoho,原来读取的是配置文件中 hibernate.id.new_generator_mappings 的值

刚好我这里的该参数的值是true

hibernate.id.new_generator_mappings=true

原来如此, 仅需把该参数的值设为false 即可解决问题

5. hibernate.id.new_generator_mappings 是什么鬼?

知其然而知其所以然,是吾辈的目标

先来看看 hibernate 对这个参数的定义javadoc

	/**
	 * Setting which indicates whether or not the new {@link org.hibernate.id.IdentifierGenerator} are used
	 * for AUTO, TABLE and SEQUENCE.
	 * Default to false to keep backward compatibility.
	 */
	public static final String USE_NEW_ID_GENERATOR_MAPPINGS = "hibernate.id.new_generator_mappings";

在 http://stackoverflow.com/questions/25047226/jpa-generationtype-auto-not-considering-column-with-auto-increment#25052275 这个文章里面, 我们可以发现点有价值的内容

写道

If you use the new identifier generators:

properties.put("hibernate.id.new_generator_mappings", "true");

The AUTO will actually use a SequenceStyleGenerator and where the database doesn't support sequences, you end up using a TABLE generator instead (which is a portable solution but it's less efficient than IDENTITY or SEQUENCE).

我们再来看看 SequenceStyleGenerator

Generator Description

SequenceStyleGenerator

It’s an enhanced version of the previous sequence generator. It uses a sequence if the underlying database supports them. If the current database doesn’t support sequences it switches to using a table for generating sequence values. While the previous generators were having a predefined optimization algorithm, the enhanced generators can be configured with an optimizer strategy:

none: there is no optimizing strategy applied, so every identifier is fetched from the database
hi/lo: it uses the original hi/lo algorithm. This strategy makes it difficult for other systems to share the same identifier sequence, requiring other systems to implement the same identifier generation logic.
pooled: This optimizer uses a hi/lo optimization strategy, but instead of saving the current hi value it stores the current range upper boundary (or lower boundary –hibernate.id.optimizer.pooled.prefer_lo).

Pooled is the default optimizer strategy.

大意是, SequenceStyleGenerator 是个增强的 sequence 生成器 ,如果底层数据库支持序列,那么使用序列, 如果不支持, 那么切换成 table来生成序列 values.

这也是,我在sqlserver 中,发现了一个 hibernate_sequence 莫名其妙的表的原因