看Hibernate源码 003 - ID Generator

Hibernate的id生成有N种策略, 可以通过hbm文件或者annotation配置.

支持的策略包括:uuid, hilo, assigned, identity, select, sequence, seqhilo, increment, foreign, guid, uuid.hex, sequence-identity.

对应这些策略, 可以在org.hibernate.id包下找到, 其中有一个IdentifierGeneratorFactory类用于根据实体类的配置(hbm文件的<id>元素或@Id,@GeneratedValue注解)来创建相应的策略.

public final class IdentifierGeneratorFactory {
... //注册所有支持的ID生成策略
private static final HashMap GENERATORS = new HashMap();
static {
   GENERATORS.put( "uuid", UUIDHexGenerator.class );
   GENERATORS.put( "hilo", TableHiLoGenerator.class );
   GENERATORS.put( "assigned", Assigned.class );
   GENERATORS.put( "identity", IdentityGenerator.class );
   GENERATORS.put( "select", SelectGenerator.class );
   GENERATORS.put( "sequence", SequenceGenerator.class );
   GENERATORS.put( "seqhilo", SequenceHiLoGenerator.class );
   GENERATORS.put( "increment", IncrementGenerator.class );
   GENERATORS.put( "foreign", ForeignGenerator.class );
   GENERATORS.put( "guid", GUIDGenerator.class );
   GENERATORS.put( "uuid.hex", UUIDHexGenerator.class ); // uuid.hex is deprecated
   GENERATORS.put( "sequence-identity", SequenceIdentityGenerator.class );
}
...
public static IdentifierGenerator create (String strategy, Type type, Properties params, Dialect dialect)
    throws MappingException {
   try {
    Class clazz = getIdentifierGeneratorClass( strategy, dialect );
...
}

public static Class getIdentifierGeneratorClass(String strategy, Dialect dialect) {
   Class clazz = ( Class ) GENERATORS .get( strategy );
...
}
}
显然create()方法是用于创建ID生成器的, 而且用到了参数strategy和dialect. Hibernate在初始化SessionFactory的时候就会准备这些ID生成器. 见以下代码

SessionFactoryImpl(){
...
   Iterator classes = cfg.getClassMappings();
   while ( classes.hasNext() ) {
    PersistentClass model = (PersistentClass) classes.next();
    if ( !model.isInherited() ) {
     IdentifierGenerator generator = model.getIdentifier().createIdentifierGenerator (
       settings.getDialect(),
            settings.getDefaultCatalogName(),
            settings.getDefaultSchemaName(),
            (RootClass) model
      );
     identifierGenerators.put( model.getEntityName(), generator );
    }
   }

model.getIdentifier().createIdentifierGenerator() 最终会引用到
return IdentifierGeneratorFactory.create(
     identifierGeneratorStrategy,
     getType(),
     params,
     dialect
    );

看来SessionFactory做了太多的工作. 为了能对这么多中ID生成策略有最直观的了解, 下面给出一个各种ID策略的类关系图.

 

该图分为两部分, 蓝色先左边的ID生成器是不需要借助数据库, 采用本地算法生成的, 而右边的ID生成器则需要借助数据库提供的ID生成能力.

最主要的接口有4个, 图中用蓝色标出.
* IdentifierGenerator: ID生成器接口, 只有一个generate方法, 返回生成的ID.
* PostInsertIdentifierGenerator: 针对MySQL, MSSQL这类能为主键设置自动增长的ID生成器
* PersistentIdentifierGenerator: 针对需要访问DB来生成ID的生成器, 比如GUID.
* Configurable: 用于读取配置信息

具体每个类如何处理:
1) uuid: 是采用128位的算法生成惟一值,支持大部分的数据库
   public Serializable generate(SessionImplementor session, Object obj) {
   return new StringBuffer(36)
    .append( format( getIP () ) ).append(sep)
    .append( format( getJVM () ) ).append(sep)
    .append( format( getHiTime () ) ).append(sep)
    .append( format( getLoTime () ) ).append(sep)
    .append( format( getCount () ) ) //注: 每次加1, JVM内唯一, 通过synchronized来保证实现
    .toString();
}
protected String format(int intval) {
   String formatted = Integer.toHexString(intval);
   StringBuffer buf = new StringBuffer("00000000");
   buf.replace( 8-formatted.length(), 8, formatted );
   return buf.toString();
}
protected String format(short shortval) {
   String formatted = Integer.toHexString(shortval);
   StringBuffer buf = new StringBuffer("0000");
   buf.replace( 4-formatted.length(), 4, formatted );
   return buf.toString();

2)GUID: 通过使用数据库本身的uuid算法来实现
public class GUIDGenerator implements IdentifierGenerator {
   public Serializable generate(SessionImplementor session, Object obj)
   throws HibernateException {
       final String sql = session.getFactory().getDialect().getSelectGUIDString();
   ...
}
假如getDialect()返回的是MySQLDialect, 则返回的是
public String getSelectGUIDString() {
   return "select uuid()";
}
但是如果不支持uuid的数据库, 则抛出异常
public String getSelectGUIDString() {
   throw new UnsupportedOperationException( "dialect does not support GUIDs" );
}

3)increment:
public class IncrementGenerator implements IdentifierGenerator, Configurable {
...
public synchronized Serializable generate (SessionImplementor session, Object object)
throws HibernateException {

   if (sql!=null) {
    getNext( session ); //注:使用了一个sql: "select max(" + column + ") from " + buf.toString();
   }
   return IdentifierGeneratorFactory.createNumber(next++, returnClass);
}
}
留意这里对generate方法使用了同步, 可想如果所有ID都通过hibernate创建, 则是安全的...

4) foreign key 简而言之, 就是要取到关联的对象的ID
foreign-key的配置稍微繁琐一点, 附上一个例子: 对于帖子(Post)和评论(Comment), Comment表有一个外键fk_post, comment.post_id关联到Post.id. 那么在Comment.hbm.xml中就可以这样配置

Comment.hbm.xml
------------------------------------------------------------
<hibernate-mapping package="work">
    <class name="Comment" lazy="false">
        <id name="id">
            <generator class="foreign">
                <param name="property">post</param>
            </generator>
        </id>
...
        <many-to-one name="post" column="post_id"></many-to-one>
    </class>
</hibernate-mapping>


hibernate源代码:
------------------------------------------------------------
public Serializable generate (SessionImplementor sessionImplementor, Object object )
throws HibernateException {
    //注:这里object是Comment对象
  
   Session session = (Session) sessionImplementor;

   //注:这里associatedObject是Post对象
   Object associatedObject = sessionImplementor.getFactory()
          .getClassMetadata( entityName )
          .getPropertyValue( object , propertyName, session.getEntityMode() );
  
   if ( associatedObject == null ) {
    throw new IdentifierGenerationException(
      "attempted to assign id from null one-to-one property: " +
      propertyName
     );
   }
  
   EntityType type = (EntityType) sessionImplementor.getFactory()
        .getClassMetadata( entityName )
        .getPropertyType( propertyName );

   Serializable id;
   try {
    id = ForeignKeys.getEntityIdentifierIfNotUnsaved(
      type.getAssociatedEntityName(),
     associatedObject ,
      sessionImplementor
     );
   }
   catch (TransientObjectException toe) {
    id = session.save( type.getAssociatedEntityName(), associatedObject );
   //注: 尝试保存该对象来生成ID, 这个操作可能触发一系列其他的东西, 如事件, 缓存写入等等
   }

   if ( session.contains(object) ) {
    //abort the save (the object is already saved by a circular cascade)
    return IdentifierGeneratorFactory.SHORT_CIRCUIT_INDICATOR;
    //throw new IdentifierGenerationException("save associated object first, or disable cascade for inverse association");
   }
   return id;
}

5) Identity: 利用数据库的自增长方式来生成ID
相比前面的策略, 这是很有意思的ID生成策略, 因为hibernate并不能在insert前预先获得ID, 而是在insert后, 依赖于JDBC API的PreparedStatement.getGeneratedKeys()方法来取得ID, 该方法返回的是一个ResultSet, 只有一列, 名称为GENERATED_KEY. 所以Hibernate也是采用一种后置处理的方式: 即在调用到IdentifierGenerator.getnerate()方法的时候(其实这个时候的实现是IdentityGenerator类) , 直接返回一个Serilizable对象--IdentifierGeneratorFactory.POST_INSERT_INDICATOR. 接着, 在我们使用session.save(object)方法的时候, 会判断save操作的类型是否为IdentifierGeneratorFactory.POST_INSERT_INDICATOR,再进行相应处理.

save部分代码量太大, 免了. 看一些关键的.

先是"奇怪"的generate()方法
public abstract class AbstractPostInsertGenerator implements PostInsertIdentifierGenerator {
public Serializable generate(SessionImplementor s, Object obj) {
   return IdentifierGeneratorFactory.POST_INSERT_INDICATOR ;
}
}
public class IdentityGenerator extends AbstractPostInsertGenerator {
.. //没有覆盖generate()
}

然后是session.save()对应的事件监听器 AbstractSaveEventListener 的saveWithGeneratedId()
protected Serializable saveWithGeneratedId(...){
...
if ( generatedId == null ) {
    throw new IdentifierGenerationException( "null id generated for:" + entity.getClass() );
   }
   else if ( generatedId == IdentifierGeneratorFactory.SHORT_CIRCUIT_INDICATOR ) {
    return source.getIdentifier( entity );
   }
   else if ( generatedId == IdentifierGeneratorFactory.POST_INSERT_INDICATOR ) {
    return performSave( entity, null, persister, true, anything, source, requiresImmediateIdAccess );
   }
...
}
该方法一直执行到protected Serializable performSaveOrReplicate(...)方法的
if ( useIdentityColumn ) {
    EntityIdentityInsertAction insert = new EntityIdentityInsertAction(
      values, entity, persister, source, shouldDelayIdentityInserts
    );
    if ( !shouldDelayIdentityInserts ) {
     log.debug( "executing identity-insert immediately" );
     source.getActionQueue().execute( insert ); //这里有文章,hibernate已先把先前的操作先转换成sql执行

经过N多处理后, 最后回到刚才提到的JDBC API上. 在IdentityGenerator.GetGeneratedKeysDelegate子类中
   public Serializable executeAndExtract (PreparedStatement insert ) throws SQLException {
    insert.executeUpdate();
    ResultSet rs = null;
    try {
     rs = insert.getGeneratedKeys ();
     return IdentifierGeneratorFactory.getGeneratedIdentity(rs,persister.getIdentifierType());
    ...  
}
~_~

不得不佩服Hibernate团队的重构能力, 许多功能都被按照"行为向上集中"的规则处理, 即相同的行为, 放在更高层次的父类去. 调式过程中能看到很多类都是继承于AbstractXXX. 并且每个类的责任很明确. 像EntityPersister, Batcher, ActionQueue, Executable和它的实现类, 等等...