Making a OneToOne-relation lazy

Question

In this application we are developing, we noticed that a view was particularly slow. I profiled the view and noticed that there was one query executed by hibernate which took 10 seconds even if there only were two object in the database to fetch. All OneToMany and ManyToMany relations were lazy so that wasn't the problem. When inspecting the actual SQL being executed, I noticed that there were over 80 joins in the query.

Further inspecting the issue, I noticed that the problem was caused by the deep hierarchy of OneToOne and ManyToOne relations between entity classes. So, I thought, I'll just make them fetched lazy, that should solve the problem. But annotating either @OneToOne(fetch=FetchType.LAZY) or @ManyToOne(fetch=FetchType.LAZY) doesn't seem to work. Either I get an exception or then they are not actually replaced with a proxy object and thus being lazy.

Any ideas how I'll get this to work? Note that I do not use the persistence.xml to define relations or configuration details, everything is done in java code.

Answer 1

To get lazy loading working on nullable one-to-one mappings you need to let hibernate do compile time instrumentation and add a @LazyToOne(value = LazyToOneOption.NO_PROXY) to the one-to-one relation.

Example Mapping:

@OneToOne(fetch = FetchType.LAZY)  
@JoinColumn(name="other_entity_fk")
@LazyToOne(value = LazyToOneOption.NO_PROXY)
public OtherEntity getOther()

Example Ant Build file extension (for doing the Hibernate compile time instrumentation):

<property name="src" value="/your/src/directory"/><!-- path of the source files --> 
<property name="libs" value="/your/libs/directory"/><!-- path of your libraries --> 
<property name="destination" value="/your/build/directory"/><!-- path of your build directory --> 

<fileset id="applibs" dir="${libs}"> 
  <include name="hibernate3.jar" /> 
  <!-- include any other libraries you'll need here --> 
</fileset> 

<target name="compile"> 
  <javac srcdir="${src}" destdir="${destination}" debug="yes"> 
    <classpath> 
      <fileset refid="applibs"/> 
    </classpath> 
  </javac> 
</target> 

<target name="instrument" depends="compile"> 
  <taskdef name="instrument" classname="org.hibernate.tool.instrument.javassist.InstrumentTask"> 
    <classpath> 
      <fileset refid="applibs"/> 
    </classpath> 
  </taskdef> 

  <instrument verbose="true"> 
    <fileset dir="${destination}"> 
      <!-- substitute the package where you keep your domain objs --> 
      <include name="/com/mycompany/domainobjects/*.class"/> 
    </fileset> 
  </instrument> 
</target>

Answer 2

First off, some clarifications to KLE's answer:

1. Unconstrained (nullable) one-to-one association is the only one that can not be proxied without bytecode instrumentation. The reason for this is that owner entity MUST know whether association property should contain a proxy object or NULL and it can't determine that by looking at its base table's columns due to one-to-one normally being mapped via shared PK, so it has to be eagerly fetched anyway making proxy pointless. Here's a more detailed explanation.

2. many-to-one associations (and one-to-many, obviously) do not suffer from this issue. Owner entity can easily check its own FK (and in case of one-to-many, empty collection proxy is created initially and populated on demand), so the association can be lazy.

3. Replacing one-to-one with one-to-many is pretty much never a good idea. You can replace it with unique many-to-one but there are other (possibly better) options.

Rob H. has a valid point, however you may not be able to implement it depending on your model (e.g. if your one-to-one association is nullable).

Now, as far as original question goes:

A) @ManyToOne(fetch=FetchType.LAZY) should work just fine. Are you sure it's not being overwritten in the query itself? It's possible to specify join fetch in HQL and / or explicitly set fetch mode via Criteria API which would take precedence over class annotation. If that's not the case and you're still having problems, please post your classes, query and resulting SQL for more to-the-point conversation.

B) @OneToOne is trickier. If it's definitely not nullable, go with Rob H.'s suggestion and specify it as such:

@OneToOne(optional = false, fetch = FetchType.LAZY)

Otherwise, if you can change your database (add a foreign key column to owner table), do so and map it as "joined":

@OneToOne(fetch = FetchType.LAZY)
@JoinColumn(name="other_entity_fk")
public OtherEntity getOther()

and in OtherEntity:

@OneToOne(mappedBy = "other")
public OwnerEntity getOwner()

If you can't do that (and can't live with eager fetching) bytecode instrumentation is your only option. I have to agree with CPerkins, however - if you have 80!!! joins due to eager OneToOne associations, you've got bigger problems then this :-)

Answer 3

Here's something that has been working for me (without instrumentation):

Instead of using @OneToOne on both sides, I use @OneToMany in the inverse part of the relationship (the one with mappedBy). That makes the property a collection (List in the example below), but I translate it into an item in the getter, making it transparent to the clients.

This setup works lazily, that is, the selects are only made when getPrevious() or getNext() are called - and only one select for each call.

The table structure:

CREATE TABLE `TB_ISSUE` (
    `ID`            INT(9) NOT NULL AUTO_INCREMENT,
    `NAME`          VARCHAR(255) NULL,
    `PREVIOUS`      DECIMAL(9,2) NULL
    CONSTRAINT `PK_ISSUE` PRIMARY KEY (`ID`)
);
ALTER TABLE `TB_ISSUE` ADD CONSTRAINT `FK_ISSUE_ISSUE_PREVIOUS`
                 FOREIGN KEY (`PREVIOUS`) REFERENCES `TB_ISSUE` (`ID`);

The class:

@Entity
@Table(name = "TB_ISSUE") 
public class Issue {

    @Id
    @GeneratedValue(strategy = GenerationType.IDENTITY)
    protected Integer id;

    @Column
    private String name;

    @OneToOne(fetch=FetchType.LAZY)  // one to one, as expected
    @JoinColumn(name="previous")
    private Issue previous;

    // use @OneToMany instead of @OneToOne to "fake" the lazy loading
    @OneToMany(mappedBy="previous", fetch=FetchType.LAZY)
    // notice the type isnt Issue, but a collection (that will have 0 or 1 items)
    private List<Issue> next;

    public Integer getId() { return id; }
    public String getName() { return name; }

    public Issue getPrevious() { return previous; }
    // in the getter, transform the collection into an Issue for the clients
    public Issue getNext() { return next.isEmpty() ? null : next.get(0); }

}

Answer 4

In native Hibernate XML mappings, you can accomplish this by declaring a one-to-one mapping with the constrained attribute set to true. I am not sure what the Hibernate/JPA annotation equivalent of that is, and a quick search of the doc provided no answer, but hopefully that gives you a lead to go on.

Answer 5

As already perfectly explained by ChssPly76, Hibernate's proxies don't help with unconstrained (nullable) one-to-one associations, BUT there is a trick explained here to avoid to set up instrumentation. The idea is to fool Hibernate that the entity class which we want to use has been already instrumented: you instrument it manually in the source code. It's easy! I've implemented it with CGLib as bytecode provider and it works (ensure that you configure lazy="no-proxy" and fetch="select", not "join", in your HBM).

I think this is a good alternative to real (I mean automatic) instrumentation when you have just one one-to-one nullable relation that you want to make lazy. The main drawback is that the solution depends on the bytecode provider you are using, so comment your class accurately because you could have to change the bytecode provider in the future; of course, you are also modifying your model bean for a technical reason and this is not fine.

Answer 6

The basic idea behing the XToOnes in Hibernate is that they are not lazy in most case.

One reason is that, when Hibernate have to decide to put a proxy (with the id) or a null,
it has to look into the other table anyway to join. The cost of accessing the other table in the database is significant, so it might as well fetch the data for that table at that moment (non-lazy behaviour), instead of fetching that in a later request that would require a second access to the same table.

Edited: for details, please refer to ChssPly76 's answer. This one is less accurate and detailed, it has nothing to offer. Thanks ChssPly76.