6 Understanding Descriptors

EclipseLink uses descriptors to store the information that describes how an instance of a particular class can be represented by a data source. Descriptors own mappings that associate class instance variables with a data source and transformation routines that are used to store and retrieve values. As such, the descriptor acts as the connection between a Java object and its data source representation.

This chapter includes the following sections:

6.1 Common Descriptor Concepts

The following sections describe the concepts that are common to Object-Relational and Object-XML descriptors.

6.1.1 Descriptor Architecture

A descriptor stores all the information describing how an instance of a particular object class can be represented in a data source.

EclipseLink descriptors may contain the following information:

  • The persistent Java class it describes and the corresponding data source (database tables or XML complex type interaction)

  • A collection of mappings, which describe how the attributes and relationships for that class are represented in the data source

  • The primary key information (or equivalent) of the data source

  • A list of query keys (or aliases) for field names

  • Information for sequence numbers

  • A set of optional properties for tailoring the behavior of the descriptor, including support for caching refresh options, identity maps, optimistic locking, the event manager, and the query manager

There is a descriptor type for each data source type that EclipseLink supports. In some cases, multiple descriptor types are valid for the same data source type. The type of descriptor you use determines the type of mappings that you can define.

6.1.2 Descriptors and Inheritance

Inheritance describes how a derived (child) class inherits the characteristics of its superclass (parent). You can use descriptors to describe the inheritance relationships between classes in relational and XML projects.

In the descriptor for a child class, you can override mappings that have been specified in the descriptor for a parent class, or map attributes that have not been mapped at all in the parent class descriptor.

Figure 6-1 illustrates the Vehicle object model–a typical Java inheritance hierarchy. The root class Vehicle contains two branch classes: FueledVehicle and NonFueledVehicle. Each branch class contains a leaf class: Car and Bicycle, respectively.

Figure 6-1 Example Inheritance Hierarchy

Description of Figure 6-1 follows
Description of "Figure 6-1 Example Inheritance Hierarchy "

EclipseLink recognizes the following three types of classes in an inheritance hierarchy:

  1. The root class stores information about all instantiable classes in its subclass hierarchy. By default, queries performed on the root class return instances of the root class and its instantiable subclasses. However, the root class can be configured so queries on it return only instances of itself, without instances of its subclasses.

    For example, the Vehicle class in Figure 6-1 is a root class.

  2. Branch classes have a persistent superclass and also have subclasses. By default, queries performed on the branch class return instances of the branch class and any of its subclasses. However, as with the root class, the branch class can be configured so queries on it return only instances of itself without instances of its subclasses.

    For example, the FueledVehicle class in Figure 6-1 is a branch class.

  3. Leaf classes have a persistent superclass in the hierarchy but do not have subclasses. Queries performed on the leaf class can only return instances of the leaf class.

    For example, the Car class in Figure 6-1 is a leaf class.

In the descriptor for a child class, you can override mappings that have been specified in the descriptor for a parent class, or map attributes that have not been mapped at all in the parent class descriptor.

This section includes information on the following topics: Specifying a Class Indicator

When configuring inheritance, you configure the root class descriptor with the means of determining which subclasses it should instantiate.

You can do this in one of the following ways:


All leaf classes in the hierarchy must have a class indicator and they must have the same type of class indicator (field or class extraction method). Using Class Indicator Fields

You can use a persistent attribute of a class to indicate which subclass should be instantiated. For example, in a relational descriptor, you can use a class indicator field in the root class table. The indicator field should not have an associated direct mapping unless it is set to read-only.


If the indicator field is part of the primary key, define a write-only transformation mapping for the indicator field.

You can use strings or numbers as values in the class indicator field.

The root class descriptor must specify how the value in the class indicator field translates into the class to be instantiated.

One approach is to configure the root class descriptor with a class indicator dictionary: a collection of key-values that associates a simple key, stored in the class indicator field, with a class to instantiate. Table 6-1 illustrates the class indicator dictionary for the Vehicle class' subclasses, as shown in Figure 6-1.

Table 6-1 Class Indicator Dictionary for the Vehicle Class

Key Value









Another approach is to simply use the class name itself as the value stored in the class indicator field. This avoids having to define unique indicators for each class at the expense of a slightly larger key value (depending on the length of your class names). Using Class Extraction Methods

You can define a Java method to compute the class indicator based on any available information in the object's data source record. Such a method is called a class extraction method.

Using a class extraction method, you do not need to include an explicit class indicator field in your data model and you can handle relationships that are too complex to describe using class indicator fields.

A class extraction method must have the following characteristics:

  • it must be defined on the root descriptor's class;

  • it must be static;

  • it must take a Record as an argument;

  • it must return the java.lang.Class object to use for the Record passed in.

You may also need to define only-instances and with-all-subclasses expressions. If you use a class extraction method, then you must provide EclipseLink with expressions to correctly filter sibling instances for all classes that share a common table.

For example, Table 6-2 lists the rows in the EMPLOYEE table. The Employee class is the base class. Director, Manager, Programmer, and TechWriter classes each derive from the Employee class. However, in your application, instances of Manager, Programmer, and TechWriter classes must be represented as Employee instances and instances of Director must be represented as Director instances. Because there is not a one-to-one correspondence between class and JOB_TYPE field value, the JOB_TYPE field alone cannot serve as a class indicator field (see Section, "Using Class Indicator Fields"). To resolve this issue, you could use the class extraction method, shown in Example 6-1.

Table 6-2 EMPLOYEE Table



Bob Jones




Sarah Smith


Technical Writer


Ben Ng




Sally Johnson



Example 6-1 Class Extraction Method

// If the JOB_TYPE field value in record equals 2, return the Director class.
// Return the Employee class for all other JOB_TYPE field values

public static Class getClassFromRecord(Record record) {
    if (record.get("JOB_TYPE").equals(new Integer(2)) {
        return Director.class;
    else {
        return Employee.class;

When configuring inheritance using a class extraction method, EclipseLink does not generate SQL for queries on the root class. Inheritance and Primary Keys

For relational projects, EclipseLink assumes that all of the classes in an inheritance hierarchy have the same primary key, as set in the root descriptor. Single and Multi-Table Inheritance

In a relational project, you can map your inheritance hierarchy to a single table or to multiple tables. Aggregate and Composite Descriptors and Inheritance

You can designate relational descriptors as aggregates. XML descriptors are always composites (see Section 6.1.3, "Descriptors and Aggregation").

When configuring inheritance for a relational aggregate descriptor, all the descriptors in the inheritance tree must be aggregates. The descriptors for aggregate and non-aggregate classes cannot exist in the same inheritance tree.

When configuring inheritance for an XML descriptor, because all XML descriptors are composites, descriptor type does not restrict inheritance.

6.1.3 Descriptors and Aggregation

Two objects—a source (parent or owning) object and a target (child or owned) object—are related by aggregation if there is a strict one-to-one relationship between them, and all the attributes of the target object can be retrieved from the same data source representation as the source object. This means that if the source object exists, then the target object must also exist, and if the source object is destroyed, then the target object is also destroyed.

In this case, the descriptors for the source and target objects must be designated to reflect this relationship.

In EJB 3.0, an aggregate is known as an embeddable. In the EJB 3.0 specification, an embeddable may not contain another embeddable (that is, the EJB 3.0 specification does not support nested aggregates).

For more information, see Section, "Aggregate and Composite Descriptors and Inheritance".

6.1.4 Descriptor Customization

You can customize a descriptor at run time by specifying a descriptor customizer—a Java class that implements the org.eclipse.persistence.config.DescriptorCustomizer interface and provides a default (zero-argument) constructor.

You use a descriptor customizer to customize a descriptor at run time through code API similar to how you use an amendment method to customize a descriptor. See Section 6.1.5, "Amendment and After-Load Methods".

6.1.5 Amendment and After-Load Methods

You can associate a static Java method that is called when a descriptor is loaded at run time. This method can amend the run-time descriptor instance through the descriptor Java code API.

You can only modify descriptors before the session has been connected; you should not modify descriptors after the session has been connected.

6.1.6 Descriptor Event Manager

In relational projects, EclipseLink raises various instances of DescriptorEvent during the persistence life cycle. Each descriptor owns an instance of DescriptorEventManager that is responsible for receiving these events and dispatching them to the descriptor event handlers registered with it.

Using a descriptor event handler, you can execute your own application specific logic whenever descriptor events occur, allowing you to take customized action at various points in the persistence life-cycle. For example, using a descriptor event handler, you can do the following:

  • Synchronize persistent objects with other systems, services, and frameworks

  • Maintain nonpersistent attributes of which EclipseLink is not aware

  • Notify other objects in the application when the persistent state of an object changes

  • Implement complex mappings or optimizations not directly supported by EclipseLink mappings

6.2 Object-Relational Descriptor Concepts

The following sections describe the concepts specific to Object-Relational descriptors.

6.2.1 Fetch Groups

By default, when you execute an object-level read query for a particular object class, EclipseLink returns all the persistent attributes mapped in the object's descriptor. With this single query, all the object's persistent attributes are defined, and calling their get methods returns the value directly from the object.

When you are interested in only some of the attributes of an object, it may be more efficient to return only a subset of the object's attributes using a fetch group.

Using a fetch group, you can define a subset of an object's attributes and associate the fetch group with either a ReadObjectQuery or ReadAllQuery query. When you execute the query, EclipseLink retrieves only the attributes in the fetch group. EclipseLink automatically executes a query to fetch all the attributes excluded from this subset when and if you call a get method on any one of the excluded attributes.

You can define more than one fetch group for a class. You can optionally designate at most one such fetch group as the default fetch group. If you execute either a ReadObjectQuery or ReadAllQuery query without specifying a fetch group, EclipseLink will use the default fetch group, unless you configure the query otherwise.

Before using fetch groups, Oracle recommends that you perform a careful analysis of system use. In many cases, the extra queries required to load attributes not in the fetch group could well offset the gain from the partial attribute loading.

Fetch groups can be used only with basic mappings configured with FetchType.LAZY (partial object queries).

EclipseLink uses the AttributeGroup that can be used to configure the use of partial entities in fetch, load, copy, and merge operations.

  • Fetch: Control which attributes and their associated columns are retrieved from the database

  • Load: Control which relationships in the entities returned from a query are populated

  • Copy: Control which attributes are copied into a new entity instance

  • Merge: Merge only those attributes fetched, loaded, or copied into an entity AttributeGroup Types and Operations

The following sections describe the possible AttributeGroup types and operations. FetchGroup

The FetchGroup defines which attributes should be fetched (selected from the database) when the entity is retrieved as the result of a query execution. The inclusion of relationship attributes in a FetchGroup only determines if the attribute's required columns should be fetched and populated. In the case of a lazy fetch type the inclusion of the attribute simply means that its proxy will be created to enable lazy loading when accessed. To force a relationship mapping to be populated when using a FetchGroup on a query the attribute must be included in the group and must either be FetchType.EAGER or it must be included in an associated LoadGroup on the query. Default FetchGroup

FetchGroup also has the notion of named and default FetchGroup which are managed by the FetchGroupManager. A default FetchGroup is defined during metadata processing if one or more basic mappings are configured to be lazy and the entity class implements FetchGroupTracker (typically introduced through weaving). The default FetchGroup is used on all queries for this entity type where no explicit FetchGroup or named FetchGroup is configured. Named FetchGroup

A Named FetchGroup can be defined for an entity using @FetchGroup annotation or within the eclipselink-orm.xml file. Full FetchGroup

A FetchGroup when first created is assumed to be empty. The user must add the attributes to the FetchGroup. If a FetchGroup is required with all of the attributes then the FetchGroupManager.createFullFetchGroup() must be used. Load/LoadAll with FetchGroup

A FetchGroup can also be configured to perform a load operation of relationship mappings and nested relationship mappings. LoadGroup

A LoadGroup is used to force a specified set of relationship attributes to be populated in a query result. CopyGroup

The CopyGroup replaces the deprecated ObjectCopyPolicy being used to define how a entity is copied. In addition to specifying the attributes defining what should be copied from the source entity graph into the target copy the CopyGroup also allows definition of:

  • shouldResetPrimaryKey: Reset the identifier attributes to their default value. This is used when the copy operation is intended to clone the entity in order to make a new entity with similar state to the source. Default is false.

  • shouldResetVersion: Reset the optimistic version locking attribute to its default value in the copies. Default is false.

  • depth: defines cascade mode for handling relationships. By default CASCADE_PRIVATE_PARTS is used but it can also be configured to NO_CASCADE and CASCADE_ALL_PARTS. Merging

When a partial entity is merged into a persistence context that has an AttributeGroup associated with it defining which attributes are available only those attributes are merged. The relationship mappings within the entity are still merged according to their cascade merge settings.

6.2.2 Descriptor Query Manager

Each relational descriptor provides an instance of DescriptorQueryManager that you can use to configure the following:

  • named queries

  • custom default queries for basic persistence operations

  • additional join expressions

6.2.3 Descriptors and Sequencing

An essential part of maintaining object identity is managing the assignment of unique values (that is, a specific sequence) to distinguish one object instance from another.

Sequencing options you configure at the project (or session) level determine the type of sequencing that EclipseLink uses. In a POJO project, you can use session-level sequence configuration to override project-level sequence configuration, on a session-by-session basis, if required.

After configuring the sequence type, for each descriptor's reference class, you must associate one attribute, typically the attribute used as the primary key, with its own sequence.

6.2.4 Descriptors and Locking

With object-relational mapping, you can configure a descriptor with any of the following locking policies to control concurrent access to a domain object:

  • Optimistic—All users have read access to the data. When a user attempts to make a change, the application checks to ensure the data has not changed since the user read the data.

  • Pessimistic—The first user who accesses the data with the purpose of updating it locks the data until completing the update.

  • No locking—The application does not prevent users overwriting each other's changes.

Oracle recommends using optimistic locking for most types of applications to ensure that users do not overwrite each other's changes.

This section describes the various types of locking policies that EclipseLink supports, including the following: Optimistic Version Locking Policies

With optimistic locking, all users have read access to the data. When a user attempts to make a change, the application checks to ensure the data has not changed since the user read the data.

Optimistic version locking policies enforce optimistic locking by using a version field (also known as a write-lock field) that you provide in the reference class that EclipseLink updates each time an object change is committed.

EclipseLink caches the value of this version field as it reads an object from the data source. When the client attempts to write the object, EclipseLink compares the cached version value with the current version value in the data source in the following way:

  • If the values are the same, EclipseLink updates the version field in the object and commits the changes to the data source.

  • If the values are different, the write operation is disallowed because another client must have updated the object since this client initially read it.

EclipseLink provides the following version-based optimistic locking policies:

  • VersionLockingPolicy

  • TimestampLockingPolicy

For descriptions of these locking policies, see "Setting Optimistic Locking" in Oracle Fusion Middleware Solutions Guide for Oracle TopLink.


In general, Oracle recommends numeric version locking because of the following:

  • accessing the timestamp from the data source can have a negative impact on performance;

  • time stamp locking is limited to the precision that the database stores for timestamps.

Whenever any update fails because optimistic locking has been violated, EclipseLink throws an OptimisticLockException. This should be handled by the application when performing any database modification. The application must notify the client of the locking contention, refresh the object, and have the client reapply its changes.

You can choose to store the version value in the object as a mapped attribute, or in the cache. In three-tier applications, you typically store the version value in the object to ensure it is passed to the client when updated (see Section, "Applying Locking in an Application").

If you store the version value in the cache, you do not need to map it. If you do map the version field, you must configure the mapping as read-only.

To ensure that the parent object's version field is updated whenever a privately owned child object is modified, consider Section, "Optimistic Version Locking Policies and Cascading".

If you are using a stored procedure to update or delete an object, your database may not return the row-count required to detect an optimistic lock failure, so your stored procedure is responsible for checking the optimistic lock version and throwing an error if they do not match. Only version locking is directly supported with a StoredProcedureCall. Because timestamp and field locking require two versions of the same field to be passed to the call, an SQL call that uses an ## parameter to access the translation row could be used for other locking policies. Optimistic Version Locking Policies and Cascading

If your database schema is such that both a parent object and its privately owned child object are stored in the same table, then if you update the child object, the parent object's version field will be updated.

However, if the parent and its privately owned child are stored in separate tables, then changing the child will not, by default, update the parent's version field.

To ensure that the parent object's version field is updated in this case, you can either manually update the parent object's version field or, if you are using a VersionLockingPolicy, you can configure EclipseLink to automatically cascade the child object's version field update to the parent.

After you enable optimistic version locking cascading, when a privately owned child object is modified, EclipseLink will traverse the privately owned foreign reference mappings, updating all the parent objects back to the root.

Optimistic version locking cascading is only applied if the child object is registered in a unit of work.

EclipseLink supports optimistic version locking cascading for:

  • object changes in privately owned one-to-one and one-to-many mappings

  • relationship changes (adding or removing) in the following collection mappings (privately owned or not):

    • direct collection

    • one-to-many

    • many-to-many

    • aggregate collection

Consider the example object graph shown in Figure 6-2

Figure 6-2 Optimistic Version Locking Policies and Cascading Example

Description of Figure 6-2 follows
Description of "Figure 6-2 Optimistic Version Locking Policies and Cascading Example"

In this example, ObjectA privately owns ObjectB, and ObjectB privately owns ObjectC, and ObjectC privately owns ObjectD.

Suppose you register ObjectB in a unit of work, modify an ObjectB field, and commit the unit of work. In this case, ObjectB checks the cache for ObjectA and, if not present, queries the database for ObjectA. ObjectB then notifies ObjectA of its change. ObjectA forces an update on its version optimistic locking field even though it has no changes to its corresponding table.

Suppose you register ObjectA in a unit of work, access its ObjectB to access its ObjectC to access its ObjectD, modify an ObjectD field, and commit the unit of work. In this case, ObjectD notifies ObjectC of its changes. ObjectC forces an update on its version optimistic locking field even though it has no changes to its corresponding table. ObjectC then notifies ObjectB of the ObjectD change. ObjectB then notifies ObjectA of the ObjectD change. ObjectA forces an update on its version optimistic locking field even though it has no changes to its corresponding table. Optimistic Locking and Rollbacks

With optimistic locking, use the UnitOfWork method commitAndResumeOnFailure to rollback a locked object's value, if you store the optimistic lock versions in the cache.

If you store the locked versions in an object, you must refresh the objects (or their versions) on a failure. Alternatively, you can acquire a new unit of work on the failure and reapply any changes into the new unit of work. Optimistic Field Locking Policies

Optimistic field locking policies enforce optimistic locking by using one or more of the fields that currently exist in the table to determine if the object has changed since the client read the object.

The unit of work caches the original state of the object when you first read the object or register it with the unit of work. At commit time, the unit of work compares the original values of the lock fields with their current values on the data source during the update. If any of the lock fields' values have changed, an optimistic lock exception is thrown.

EclipseLink provides the following optimistic field locking policies:

  • AllFieldsLockingPolicy

  • ChangedFieldsLockingPolicy

  • SelectedFieldsLockingPolicy

  • VersionLockingPolicy

  • TimestampLockingPolicy

For descriptions of these locking policies, see "Setting Optimistic Locking" in Oracle Fusion Middleware Solutions Guide for Oracle TopLink. Pessimistic Locking Policies

With pessimistic locking, the first user who accesses the data with the purpose of updating it locks the data until completing the update.

When using a pessimistic locking policy, you can configure the policy to either fail immediately or to wait until the read lock is acquired.

You can use a pessimistic locking policy only in a project with a container-managed persistence type and with descriptors that have EJB information.

You can also use pessimistic locking (but not a pessimistic locking policy) at the query level.

EclipseLink provides an optimization for pessimistic locking when this locking is used with entities with container-managed persistence: if you set your query to pessimistic locking and run the query in its own new transaction (which will end after the execution of the finder), then EclipseLink overrides the locking setting and does not append FOR UPDATE to the SQL. However, the use of this optimization may produce an undesirable result if the pessimistic lock query has been customized by the user with a SQL string that includes FOR UPDATE. In this case, if the conditions for the optimization are present, the query will be reset to nonpessimistic locking, but the SQL will remain the same resulting in the locking setting of the query conflicting with the query's SQL string. To avoid this problem, you can take one of the following two approaches: Applying Locking in an Application

To correctly lock an object in an application, you must obtain the lock before the object is sent to the client for editing. Applying Optimistic Locking in an Application

If you are using optimistic locking, you have the following two choices for locking objects correctly:

  1. Map the optimistic lock field in your object as not read-only and pass the version to the client on the read and back to the server on the update.

    Ensure that the original version value is sent to the client when it reads the object for the update. The client must then pass the original version value back with the update information, and this version must be set into the object to be updated after it is registered/read in the new unit of work on the server.

  2. Hold the unit of work for the duration of the interaction with the client.

    Either through a stateful session bean, or in an HTTP session, store the unit of work used to read the object for the update for the duration of the client interaction.

    You must read the object through this unit of work before passing it to the client for the update. This ensures that the version value stored in the unit of work cache or in the unit of work clone will be the original value.

    This same unit of work must be used for the update.

The first option is more commonly used, and is required if developing a stateless application. Applying Pessimistic Locking in an Application

If you are using pessimistic locking, you must use the unit of work to start a database transaction before the object is read. You must hold this unit of work and database transaction while the client is editing the object and until the client updates the object. You must use this same unit of work to update the object.

6.3 Object-XML Descriptor Concepts

The following section describes the use of default root elements in Object-XML descriptors.

6.3.1 Default Root Element

You configure XML descriptors with a default root element so that the EclipseLink runtime knows the data source data type associated with the class the descriptor describes.


The undefined document root element of a referenced object is ignored during marshalling with an any collection mapping and object mapping.

This section describes what a default root element is and how EclipseLink uses it.

Consider the Customer and Address classes and their mappings, shown in Example 6-2.

Example 6-2 Customer and Address Classes

Class: Customer
Default Root: customer
Attributes and Mappings:
    name:String                Direct Mapping                   name/text()
    billingAddress:Address     Composite Object Mapping to     billing-address
    shippingAddress:Address    Composite Object Mapping to     shipping-address

Class: Address
Default Root: address
Attributes and Mappings:
    street:String              Direct Mapping to               street/text()
    city:String                Direct Mapping to               city/text()

These classes correspond to the XML schema, shown in Example 6-3.

Example 6-3 Customer and Address Schema

<xsd:schema xmlns:xs="http://www.w3.org/2001/XMLSchema">
    <xsd:complexType name="address-type">
            <element name="street" type="xsd:string"/>
            <element name="city" type="xsd:string"/>
    <xsd:element name="customer" type="customer-type"/>
    <xsd:complexType name="customer-type">
            <xsd:element name="name" type="xsd:string"/>
            <xsd:element name="billing-address" type="address-type"/>
            <xsd:element name="shipping-address" type="address-type"/>

When an instance of the Customer class is persisted to XML, the EclipseLink runtime performs the following:

  1. Gets the default root element.

    The Customer class instance corresponds to the root of the XML document. The EclipseLink runtime uses the default root element specified on the descriptor (customer) to start the XML document. EclipseLink then uses the mappings on the descriptor to marshal the object's attributes:

  2. When the EclipseLink runtime encounters an object attribute such as billingAddress, it checks the mapping associated with it to determine with what element (billing-address) to continue:


    The EclipseLink runtime checks the mapping's reference descriptor (Address) to determine what attributes to persist:


6.4 Descriptor API

The Descriptor API can be used to define, or amend EclipseLink descriptors through Java code. The Descriptor API classes are mainly in the org.eclipse.persistence.descriptors package. These include the following classes:

  • ClassDescriptor (abstract generic Descriptor API)

  • RelationalDescriptor (relational project-specific API)

  • DescriptorEventManager (event API)

  • DescriptorQueryManager (query API)

  • InheritancePolicy

  • InterfacePolicy

  • ReturningPolicy

  • Locking policies (various optimistic locking policies)

The Descriptor API also includes these non-relational descriptors:

  • org.eclipse.persistence.oxm.XMLDescriptor

  • org.eclipse.persistence.eis.EISDescriptor

For object-relational data type and XML projects, descriptor classes are in the org.eclipse.persistence.mappings.structures and org.eclipse.persistence.oxm packages, respectively.

The following illustrates the hierarchy of descriptor types that derive from the class org.eclipse.persistence.descriptors.ClassDescriptor.

class org.eclipse.persistence.descriptors.ClassDescriptor
    class org.eclipse.persistence.descriptors.RelationalDescriptor
        class org.eclipse.persistence.mappings.structures.ObjectRelationalDataTypeDescriptor
    class org.eclipse.persistence.oxm.XMLDescriptor