Hibernate缓存命中机制[转]

通过这篇文章纪录hibernate二级缓存的一些使用经历,利用几个test case,从代码角度说明二级缓存在使用过程中一些需要注意的问题 

使用到的Model类有两个，Author, Book, 两者之间为一对多的关系 

@Entity
@Cache(usage = CacheConcurrencyStrategy.READ_WRITE)
public class Author {

    private Long id;
    private String name;

    private Set<Book> books = new HashSet<Book>();
        // getter setter methods omitted
｝

@Entity
@Cache(usage = CacheConcurrencyStrategy.READ_WRITE)
public class Book {

    private Long id;
    private String title;

    private Author author;
        // getter setter methods omitted
}

主要的测试类为TestHibernateSecondLevelCache.java 

public class TestHibernateSecondLevelCache {

    protected Logger logger = LoggerFactory.getLogger(getClass());

    private static SessionFactory sessionFactory;

    @BeforeClass
    public static void setUpSessionFactory(){
        sessionFactory = new AnnotationConfiguration().configure().buildSessionFactory();
    }

    @After
    public void clearSecondLevelCache(){
        logger.info("clear second level cache");
        sessionFactory.evict(Author.class);
        sessionFactory.evict(Book.class);
                sessionFactory.getStatistics().clear();
    }

    private Session openSession(){
        return sessionFactory.openSession();
    }

    private Statistics getStatistics(){
        return sessionFactory.getStatistics();
    }
}

方法setUpSessionFactory用于创建Hibernate SessionFactory,因为创建Session Factory是个相对比较耗时的操作,因此加上Junit4的@BeforeClass annotation,表示该Session Factory只会创建一次,被所有的test case共享.而clearSecondLevelCache方法会在每个test case结束时调用,用于清空二级缓存,防止前一个test case的結果影响后一个test case 

测试使用的hibernate-core版本为:3.3.2.GA, hibernate-annotations版本为:3.4.0.GA,测试的数据库为hsqldb内存数据库 

一. session.get() 

先来看一下session.get是否会查找二级缓存 

   @Test
public void testSessionGetCache(){
    Author author = createAuthor();

    assertGetMissCache(Author.class, author.getId());
    assertGetHitCache(Author.class, author.getId());

    updateAuthor(author);

    assertGetMissCache(Author.class, author.getId());
}

   private Author createAuthor(){
    Session session = openSession();
    Author author = new Author();
    author.setName("septem");
    session.save(author);
    session.close();
    return author;
}

@SuppressWarnings("unchecked")
private void assertGetMissCache(Class clazz, Serializable id){
    Statistics stat = getStatistics();
    long missCount = stat.getSecondLevelCacheMissCount();
    Session session = openSession();
    session.get(clazz, id);
    session.close();
    assertEquals(missCount + 1, stat.getSecondLevelCacheMissCount());
}

@SuppressWarnings("unchecked")
private void assertGetHitCache(Class clazz, Serializable id){
    Statistics stat = getStatistics();
    long hitCount = stat.getSecondLevelCacheHitCount();
    Session session = openSession();
    session.get(clazz, id);
    session.close();
    assertEquals(hitCount + 1, stat.getSecondLevelCacheHitCount());
}

   private void updateAuthor(Author author){
    author.setName("new_name");
    Session session = openSession();
    session.update(author);
    session.flush();
    session.close();
}

testSessionGetCache首先通过createAuthor创建一个author对象,然后在assertGetMissCache里面通过author.id使用get方法查出之前创建的author,因为这是每一次调用get方法,所以hibernate从数据库取回author对象,并将它存入二级缓存.测试結果通过hibernate statistics统计信息里的second level cache miss count来判断这次的get查询未命中缓存 

接着assertGetHitCache用同一个id通过get方法获取author对象,因为这个id的对象之前已存入二级缓存,所以这次操作命中缓存 

最后通过updateAuthor更新之前的author对象,hibernate会自动将该对象从二级缓存中清除,因此第三次调用get方法时没有命中缓存 

总结 : session.get方法会先中二级缓存中通过id做为key查找相应的对象,如果不存在,再发送SQL语句到数据库中查询 

二. session.load() 

第二步试一下session.load方法 

@Test
    public void testSessionLoadCache(){
        Author author = createAuthor();

        assertLoadMissCache(Author.class, author.getId());
        assertLoadHitCache(Author.class, author.getId());

        updateAuthor(author);

        assertLoadMissCache(Author.class, author.getId());
    }

        @SuppressWarnings("unchecked")
    private void assertLoadMissCache(Class clazz, Serializable id){
        Statistics stat = getStatistics();
        long missCount = stat.getSecondLevelCacheMissCount();
        Session session = openSession();
        Author author = (Author) session.load(clazz, id);
        author.getName();
        session.close();
        assertEquals(missCount + 1, stat.getSecondLevelCacheMissCount());
    }

    @SuppressWarnings("unchecked")
    private void assertLoadHitCache(Class clazz, Serializable id){
        Statistics stat = getStatistics();
        long hitCount = stat.getSecondLevelCacheHitCount();
        Session session = openSession();
        session.load(clazz, id);
        Author author = (Author) session.load(clazz, id);
        author.getName();
        session.close();
        assertEquals(hitCount + 1, stat.getSecondLevelCacheHitCount());
    }

同样的結果,每一次通过id load未命中缓存,第二次通过相同的id调用load方法命中缓存,而更新过author对象后,缓存失效,第三次查询通过数据库获取author 

有一点跟get方法不同: 

Author author = (Author) session.load(clazz, id);
        author.getName();

总结: 调用load方法的时候,hibernate一开始并没有查询二级缓存或是数据库,而是先返回一个代理对象,该对象只包含id,只有显示调用对象的非id属性时,比如author.getName(),hibernate才会去二级缓存查找,如果没命中缓存再去数据库找,数据库还找不到则抛异常.load方法会尽量推迟对象的查找工作,这是它跟get方法最大的区别. 

这两者的测试用例如下: 

@Test(expected=ObjectNotFoundException.class)
    public void testSessionLoadNonexistAuthor(){
        Session session = openSession();
        Author author = (Author) session.load(Author.class, -1L);
                assertEquals(Long.valueOf(-1), author.getId());
        author.getName();
        session.close();
    }

    @Test
    public void testSessionGetNonexistAuthor(){
        Session session = openSession();
        Author author = (Author) session.get(Author.class, -1L);
        session.close();
        assertNull(author);
    }

三. session.createQuery().list() 

@SuppressWarnings("unchecked")
    @Test
    public void testSessionList(){
        Author author = createAuthor();
        createAuthor();

        Session session = openSession();
        //hit database to select authors and populate the cache
        List<Author> authors = session.createQuery("from Author").list();
        session.close();

        assertEquals(authors.size(), getStatistics().getSecondLevelCachePutCount());

        Session session2 = openSession();
        //hit database again to select authors
        session2.createQuery("from Author").list();
        session2.close();

        assertEquals(authors.size(), getStatistics().getSecondLevelCachePutCount());

        assertGetHitCache(Author.class, author.getId());
    }

首先创建2个author对象,使用HQL : "from Author"调用list方法,这时hibernate直接从数据库查询所有的author对象,并没有从缓存中查询,但是通过list方法查出的所有author对象会存入二级缓存,这点通过getStatistics().getSecondLevelCachePutCount()可以看出来 

接着再调用list方法一次,因为此时还没找开查询缓存,list方法重新从数据查了一次.因为第一次查询已将所有的author存入缓存,所以再调用get方法时会命中缓存,assertGetHitCache通过 

总结: list方法不会从二级缓存中查找,但它从数据库中查找出来的对象会被存入cache 

四. session.createQuery().iterate() 

@SuppressWarnings("unchecked")
    @Test
    public void testSessionIterate(){
        Author author = createAuthor();
        createAuthor();

        int authorCount = 0;

        Session session = openSession();
                //hit database to get ids for all author
        Iterator<Author>  it = session.createQuery("from Author").iterate();
        while(it.hasNext()){
            Author a = it.next();
            a.getName();
            authorCount++;
        }
        session.close();
        assertEquals(authorCount, getStatistics().getEntityLoadCount());
        assertGetHitCache(Author.class, author.getId());
    }

先创建2个author对象, 通过HQL: "from Author"调用iterate方法,此时hibernate并没有查author对象,而是先从数据库查出所有author的id,控制台会输入类似以下的SQL: 

select id from author

在对iterator里面遍历的时候,会根据id一个一个地从先中缓存中查找author,没找到再访问数据库 


总结: iterate方法使用的是典型的N+1次查询,先从数据库查询出所有对象的ID,再根据ID一个一个地从二级缓存查找,二级缓存找不到再查询数据库 

五. association cache 

hibernate支持对关联进行缓存,先在Book.java加上books集合的缓存配置 

@OneToMany(mappedBy = "author", fetch = FetchType.LAZY)
    @Cache(usage = CacheConcurrencyStrategy.READ_WRITE)
    public Set<Book> getBooks() {
        return books;
    }

测试用例如下: 

@Test
    public void testAssociationCache(){
        Author author = createAuthorWith3Books();
        assertGetBooksForAuthorMissCache(author, 1);
        assertGetBooksForAuthorHitCache(author, 4);
        updateOneBookForAuthor(author);
        assertGetBooksForAuthorMissCache(author, 1);
        addNewBookForAuthor(author);
        assertGetBooksForAuthorMissCache(author, 1);
    }

    private Author createAuthorWith3Books(){
        Session session = openSession();

        Author author = new Author();
        author.setName("septem");

        Book book1 = new Book();
        book1.setTitle("book1");
        book1.setAuthor(author);

        Book book2 = new Book();
        book2.setTitle("book2");
        book2.setAuthor(author);

        Book book3 = new Book();
        book3.setTitle("book3");
        book3.setAuthor(author);

        author.getBooks().add(book1);
        author.getBooks().add(book2);
        author.getBooks().add(book3);

        session.save(book1);
        session.save(book2);
        session.save(book3);

        session.close();
        return author;
    }

    private void assertGetBooksForAuthorMissCache(Author author, long miss){
        Session session = openSession();
        Author a = (Author) session.get(Author.class, author.getId());
        long missCount = getStatistics().getSecondLevelCacheMissCount();
        a.getBooks().size();
        session.close();
        assertEquals(missCount + miss, getStatistics().getSecondLevelCacheMissCount());
    }

    private void assertGetBooksForAuthorHitCache(Author author, long hit){
        Session session = openSession();
        Author a = (Author) session.get(Author.class, author.getId());
        long hitCount = getStatistics().getSecondLevelCacheHitCount();
        a.getBooks().size();
        session.close();
        assertEquals(hitCount + hit, getStatistics().getSecondLevelCacheHitCount());
    }

    private void updateOneBookForAuthor(Author author){
        Session session = openSession();

        Author a = (Author) session.get(Author.class, author.getId());
        Book book = (Book) session.get(Book.class, a.getBooks().iterator().next().getId());
        book.setTitle("new_title");
        session.flush();

        session.close();
    }

    private void addNewBookForAuthor(Author author){
        Session session = openSession();

        Author a = (Author) session.get(Author.class, author.getId());
        Book book = new Book();
        book.setTitle("new_book");
        book.setAuthor(a);
        a.getBooks().add(book);
        session.save(book);
        session.update(a);
        session.flush();
        session.close();
    }

先创建一个author,为该author添加3个book对象.在assertGetBooksForAuthorMissCache通过author.getBooks访问关联的book集合,因为延迟加载的关系,此时并没有查询缓存也没有查询数据库,在调用a.getBooks().size()也就是访问book集合的元素时,hibernate先中缓存中查找,没有发现关联缓存,重新数据库查询,生成的SQL类似如下: 

select * from book where author_id = ?   此时statistics的missCount只增加了1,因为调用author.getBooks没有命中缓存.hibernate从数据库查询出books后,将books关联以及三个book对象都存入二级缓存. 关联的缓存是以什么样的形式存在呢?注意关联缓存没有保存books集合本身,而是保存所有book的id,假设3个book对象的id分别为1, 2, 3,则author缓存的格式类似于如下:

*---------------------------------*
|        Author Data Cache        |
|---------------------------------|
| 1 -> [ "septem" , [ 1, 2, 3 ] ] |
*---------------------------------*

第二步执行assertGetBooksForAuthorHitCache(author, 4)的时候,我们看到hitCount增加了4.因为第二次调用author.getBooks的时候,命中了关联缓存,从缓存中取回3个id,又分别用id一个一个地从二级缓存中取回3个book对象,一共命中缓存4次 

接着通过updateOneBookForAuthor(author)更新了其中的一个book对象,假设更新的是id为1的book.接着的assertGetBooksForAuthorMissCache(author, 1)方法里面missCount又增加了1.book虽然更新了,但是author.getBooks还是能命中缓存,因为book id列表还是[ 1, 2, 3 ].从缓存中取回book id列表,通过book id查找book的时候,因为id为1的book已经更新过了,它的二级缓存失效了,重新去数据库取,此时missCount增加了1,而id为2,3的book还是从二级缓存中找到的.这个方法hibernate会生成类似如下的SQL: 

select * from book where id = 1

更新其中的一个book对象不会造成关联缓存的失效,但如果更新了集合id列表的话,缓存就会失效.先通过addNewBookForAuthor为author增加一个books对象,此时books集合里面一共有4个book对象,最后的assertGetBooksForAuthorMissCache(author, 1)我们可以看到缓存失效,missCount增加了1.此时同第一次调用author.getBooks一样,hibernate生成类似如下的SQL 

select * from book where author_id = ?   总结: 关联缓存保存的是集合的id列表,而不是集合本身,关联命中缓存的时候,会根据id一个一个地先从二级缓存查找,找不到再查询数据库.更新集合中的某个对象不会造成关联缓存失效,只有改变集合的id列表才会造成缓存失效 五. 查询缓存query cache 在hibernate.cfg.xml里面加上以下配置开启查询缓存:

<property name="hibernate.cache.use_query_cache">true</property>

测试用例如下: 

@Test
    public void testQueryCache(){
        createAuthor();
        createAuthor();

        assertQueryMissCache();

        assertQueryHitCache();

        createAuthor();
        assertQueryMissCache();
    }   先做准备工作,创建两个author对象,假设它们的id分别为1,2.assertQueryMissCache里面第一次调用list方法,注意调用list前必须setCacheable(true)才会使用查询缓存,此时未命中查询缓存,hibernate从数据库查询Author对象,将此次查询存入查询缓存,同时会将查询到的author对象存入二级缓存 查询缓存并不保存查询结果集,而只是保存结果集的id,它的结构类似以下数据:

*---------------------------------------------------------------*
|                         Query Cache                           |
|---------------------------------------------------------------|
| [ ["from Author where name = ?", [ "septem"] ] -> [  1, 2 ] ] |
*---------------------------------------------------------------*

注意缓存的key与HQL,参数以及分页参数有关 

再调用assertQueryHitCache()用同样的HQL与参数重新查询Author此时会命中查询缓存,并根据结果集id一个一个地查询author对象,因为author对象之前已存入二级缓存,所以这次查询也会命中二级缓存 

查询缓存的失效比较特殊,只要查询涉及的任何一张表的数据发生变化,缓存就会失效.比如我们再创建一个Author对象,此时Author表发生了变化,原来的查询缓存就失效了 


总结: 查询缓存的key与HQL,查询参数以及分布参数有关,而且一旦查询涉及到的任何一张表的数据发生了变化,缓存就失效了,所以在生产环境中命中率较低.查询缓存保存的是结果集的id列表,而不是结果集本身,命中缓存的时候,会根据id一个一个地先从二级缓存查找,找不到再查询数据库. 

涉及到的所有代码保存在google code上 

svn checkout http://hibernate-cache-testcase.googlecode.com/svn/trunk/ hibernate-cache-testcase