Customizing Data Service Update Behavior

BEA AquaLogic Data Services Platform handles updates to relational data sources automatically. However, for any non-relational data sources, including Web services, you must provide the update logic by writing an update override class and associating it with the data service. In addition, there are times when you may want (or need) to provide custom update logic for relational data sources as well.

Any data service, logical or physical, can have an associated update override class to perform a variety of customizations.

This chapter explains how to create an update override class (the class comprising the update behavior) and when you may want to do so for relational data sources. It includes the following topics:

What is an Update Override?

An update override provides you with a mechanism for customizing or completely replacing the default update process (as discussed in How It Works: The Decomposition Process). With an update override associated with your data service, you can:

Invoke data service functions or procedures.
Execute externally defined JPDs (Java process definition) to perform workflow operations from a data service. For example, you can initiate a workflow that ties together numerous data services to accomplish distributed transactional semantics across data services that comprise non-XA-compliant data sources (such as Web services).
Validate changes before submitting them, checking or modifying the values in some way.
Invoke other resources, for example, by passing modified values to a workflow or Web service.
Execute SQL statements directly within the update plan.
Log changes to an external log file.
Perform virtually any other customization required.

An Update Override is a Java Class

In programming terms, an update override is a Java class; it is a compiled Java source code file that implements the UpdateOverride interface (<UpdateOverride>, one of the DSP APIs located in the com.bea.ld.dsmediator.update package). The UpdateOverride interface has as its sole method an empty performChange( ) method (see Listing 9-1).

As shown in Listing 9-1, the performChange() method takes a DataGraph object (passed to it by the Mediator). This object is the SDO on which your update override class will operate. The DataGraph object contains the data object, the changes to the object, and other artifacts, such as metadata (as discussed in Data Services Platform and Service Data Objects (SDOs) on page 2-2.)

As you can see from the performChange( ) method signature (Listing 9-1), it returns a Boolean value. This value serves as something of a flag to the Mediator, as follows:

How an Update Override Affects Update Processing

The performChange( ) method will be executed whenever a submit is issued for objects bound to the overridden data service.

If the object being passed in the submit() is an array of DataService objects, the array is decomposed into a list of singleton DataService objects. Some of these objects may have been added, deleted, or modified; therefore, the update override might be executed more than once (that is, once per changed object.)

In your code, you should verify that the root data object for the data graph being passed at runtime is an instance of the singleton data object bound to the data service (configured with the update override).

When Are Update Overrides Required?

You must create custom update classes to update any non-relational data sources—Web services, XML files, flat-files, and DSP procedures, for example, and for these types of scenarios:

Initiate a workflow (business process or JPD) from a DSP application.
Compute your own primary key value when adding a data object as a new record to an RDBMS.
Handle circular dependencies that arise when modifying or adding objects with mutual dependencies.

For example, your client application code is adding both a department and a manager; however, manager is also a required field of department. How can you set the department's manager field before the manager exists? As follows:

1) Add department with manager set to a temporary value
2) Add the employee manager
3) Reset the department manager to the new employee.

Once you have written and compiled the Java code that comprises the update override class, you must register the class with the data service. Update overrides can be registered on physical or logical data services: Each data service has an Override Class property that can be associated with a specific Java class file that comprises the implementation of the UpdateOverride.

At runtime, the data service executes the UpdateOverride class when it identifies it as available during the decomposition process (see Logical Data Service Update Process).

For relational sources, you may also want to use custom update classes to apply custom logic to the update process, or if an aspect of the data service design prevents automated updates, as discussed in When Are Update Overrides Required for Relational Data Sources?

When Are Update Overrides Required for Relational Data Sources?

DSP automatically updates relational data sources. However, in some cases, such as those listed Table 9-1, DSP cannot automatically update relational data sources, and requires that you provide an update override to handle update processing.

Table 9-1 Issues that Can Interfere with Automatic Relational Data Source Updates
Issue	Description, example, or recommendation
Ambiguous data lineage	The data service decomposition function cannot contain "if-then-else" constructs that provide alternate composition from lower level data services.
Transformation issue	The lineage involves a transformation other than data() or rename. For example, the following would not be supported by automatic updates: `<ACCOUNT> { sum(data($C/ACCOUNT)) }; </ACCOUNT>`
Multiple lineage for a composed property	An example of a property with more than one lineage, or data source, for a property: `<customerName>{ cat(data($C/FNAME), " ", data($WS/LAST_NAME)) }; </customerName>`
Nested matching logic issue	Typically, nested containment is expressed in XQuery using a where clause. If the query does not use a where clause to implement nesting, DSP cannot determine the foreign key-primary key association. (Nested matching logic should be expressed in a where predicate clause.) For instance, if an element of a complex type has values from more than one source (that is, a data object has fields from more than one source), the where predicate does not indicate a 1-N cardinality between the two source because the where predicate does not involve a primary key. For example, any M:N join like Orders with Payments is not usually a common join, and in this case neither Orders nor Payments would be decomposed.
Ambiguous tuple identity	Distinct-values or group-by would lead to an arbitrary tuple remaining from a set of duplicate tuples.
Redundant instance values	If the same source value instance gets projected in the SDO (or the same physical data source value), and if it is updated in the SDO, it will not be automatically decomposed.
Repeating complex type values issue	In some complex types (such as Part and Item values), the Part values may repeat and are therefore not decomposed. For example: You can determine whether a primary key is projected or derivable by knowing the cardinality between two tuples that provide the data object values. If the predicate between the tuples identifies a primary key on one side (tuple1) but not on the other side (tuple2), values from tuple1 may repeat. Tuple1 values would not be decomposed, but tuple2 values would be decomposed. If the predicate identifies that both tuples primary keys are equal, then values for both tuples would be decomposed. If two Lists of Orders occur in a data object, the predicates used to produce them may or may not make them disjointed. No attempt is made to detect this case. Updates from each instance will be decomposed as separate updates. Depending on the chosen optimistic locking strategy for the data service, the second update may or may not succeed and may overwrite changes made in the first update.
Typematch issue	If the query plan of the decomposition function has a "typematch" node, the decomposition will stop at that point for the SDO.

Developing the UpdateOverride Class

Create a new Java class file in the DSP project. (If you do not add the Java class file to the project, it must be in the classpath.) You can put the class anywhere in the application folder. For basic projects, you can simply add the class to the same directory as your data services. For larger projects, you might want to keep update classes in their own folder.

Import the appropriate DSP API and SDO DataGraph packages into the class in which you are implementing the UpdateOverride interface:

import com.bea.ld.dsmediator.update.UpdateOverride;
import commonj.sdo.DataGraph;

Your Java class declaration must implement the UpdateOverride interface, as in:

public class SpecialOrders implements UpdateOverride

Add a performChange( ) method to the class. This public method takes a DataGraph object (containing the modified data object) and returns a Boolean value. For example:

public boolean performChange(DataGraph graph)

In the body of the performChange( ) method, implement your processing logic. Your processing logic can access the changed object; instantiate new data objects;modify and submit them, or access the Mediator context's update plan and decomposition map. You can also invoke a data service procedure from within this method, or invoke a JPD.

Compile the Java source code to create the class file.
Associate the class file with a specific data service by embedding the appropriate text in the data service source code (the .ds file) or by setting the Update Override property on the data service. WebLogic Workshop is used for either approach, albeit from within two different view tabs, as follows:

Add the name of the update override class (classname only, without the .class extension) as an attribute of an empty javaUpdateExit element tag (in the pragma statement of the data service). For example:

<javaUpdateExit className="SpecialOrderUpdate"/>

Alternatively, open the Property entering the class name in the Update Override property WebLogic Workshopas the update override class Update Override for specific data service by referring to it from the data service by placing a javaUpdate element in the pragma statement of the data service.

package RTLServices;

import com.bea.ld.dsmediator.update.UpdateOverride;
import commonj.sdo.DataGraph;
import java.math.BigDecimal;
import java.math.BigInteger;
import retailer.ORDERDETAILDocument;
import retailerType.LINEITEMTYPE;
import retailerType.ORDERDETAILTYPE;

public class OrderDetailUpdate implements UpdateOverride
{ 
    public boolean performChange(DataGraph graph){
      ORDERDETAILDocument orderDocument = 
                                     (ORDERDETAILDocument) graph.getRootObject();
      ORDERDETAILTYPE order =
                            orderDocument.getORDERDETAIL().getORDERDETAILArray(0); 
      BigDecimal total = new BigDecimal(0);
      LINEITEMTYPE[] items = order.getLINEITEMArray();
      for (int y=0; y < items.length; y++) {
         BigDecimal quantity = 
                        new BigDecimal(Integer.toString(items[y].getQuantity()));
         total = total.add(quantity.multiply(items[y].getPrice()));
      }
      order.setSubTotal(total);
      order.setSalesTax(
          total.multiply(new BigDecimal(".06")).setScale(2,BigDecimal.ROUND_UP));
      order.setHandlingCharge(new BigDecimal(15));
      order.setTotalOrderAmount(
            order.getSubTotal().add(
                           order.getSalesTax().add(order.getHandlingCharge())));
      System.out.println(">>> OrderDetail.ds Exit completed");
      return true;
    }
}

In the sample class shown in Listing 9-2, an OrderDetailUpdate class implements the UpdateOverride class, and, as required by the interface, defines a performChange( ) method. Listing 9-2 demonstrates a common coding pattern for update overrides:

The submitted data graph (as changed by the client application) is passed to the performChange( ) method.
The data graph's root data object is obtained and then cast to an ORDERDETAILDocument using the variable name orderDocument.

ORDERDETAILDocument orderDocument = 
                             (ORDERDETAILDocument) graph.getRootObject();

Objects in the changed object list are accessed through the appropriate get call and index value. For example, to obtain the first such object:

ORDERDETAILTYPE order =
                    orderDocument.getORDERDETAIL().getORDERDETAILArray(0)

A processing loop iterates through the objects in the array of line items and calculates sub-totals and sales tax for each order item, adding the amounts to the order object.
Finally, the method returns true and the Mediator continues with the normal course of update processing (using the modified update plan).

Invoking Data Service Procedures from an UpdateOverride

Listing 9-3 shows an example of an update override class that invokes a data service procedure. Since UpdateOverrides are invoked locally, within the DSP server, the sample uses the typed Mediator API. As shown in Listing 9-3, several Web services operations (to create, delete, and modify a customers address) have been registered with a Data Service.

public class CustomerAddressUpdate implements UpdateOverride {
   public boolean performChange(DataGraph graph) {
     bool status = true;  // assume the best
     ChangeSummary changeSum = datagraph.getChangeSummary();
     // If no changes, do nothing. 
     if (changeSum.getChangedDataObjects().size()==0) {
       return true;
     }
// Get the DataGraph's root DataObject and cast to customer object to 
// enable getting DataGraph constituents
     CUSTOMERDocument custDoc = (CUSTOMERDocument) graph.getRootObject();
       ADDRESS[] addr = custDoc.ADDRESS().getADDRESSArray();
        int i;
        try {
          CUSTOMER custDS = CUSTOMER.getInstance(
               new InitialContext(), "RTLApp" );
// For each address in the Customer's address array, call the Web Service's
// update, delete, or create procedure as appropriate
         for( i = 0; i < addr.length; i++ ) {
              if ( changeSum.isModified( addr[ i ] ) ) {
                    custDS.invokeProcedure("modifyCustomerAddress", 
                    new Object [] {addr[ i ]} );
                }
              else if ( changeSum.isDeleted( addr[ i ] ) ) {
                    custDS.invokeProcedure("deleteCustomerAddress", 
                    new Object [] {addr[ i ]});
                }
              else if ( changeSum.isCreated( addr[ i ] ) ) {
                    custDS.invokeProcedure("createCustomerAddress", 
                    new Object [] {addr[ i ]} );
                }
                else {
                    // throw an exception for IllegalState
                    }
                }
            } // end for
        }
        catch( Exception ex ) {
            System.err.println( ex.printStackTrace() );
            throw ex;
        }
        return status;
    }
}

The example in Listing 9-3 is for a Web service running locally on the WebLogic Server instance, so it does not include basic setup code to obtain context and location. (If the Web service is not local to the WebLogic Server instance, your code must obtain an InitialContext and providing appropriate location and security properties. See Obtaining a WebLogic JNDI Context for Data Services Platform for more information about InitialContext.)

Listing 9-4 shows an update override alters the update plan in order to enforce referential integrity by removing product information from the middle of a list and adds it back at the end.

	// delete order, item, product, due to RI between ITEM and Product
	// product has to be deleted after items
    public boolean performChange(DataGraph graph)
    {
    DataServiceMediatorContext context = DataServiceMediatorContext.currentContext();
       UpdatePlan up =context.getCurrentUpdatePlan( graph, false );
       Collection dsCollection = up.getDataServiceList();
       DataServiceToUpdate ds2u = null;
       for (Iterator it=dsCollection.iterator();it.hasNext();)
       {
           ds2u = (DataServiceToUpdate)it.next();
           if (ds2u.getDataServiceName().compareTo("ld:DataServices/PRODUCT.ds") == 0 ) {
// remove product from the mid of list and add it back at the end
			up.removeContainedDataService( ds2u.getDataGraph() );
			up.addDataService(ds2u.getDataGraph(), ds2u );
           };
        }
       context.executeUpdatePlan( up );
       return false;
    }
}

Testing Submit Results

Data service updates should always be tested to ensure that changes occur as expected. You can test submits using the Test View in BEA WebLogic Workshop.

The results in Test View depend on the type of changes being made, specifically, whether you are testing read and navigate functions or DSP procedures. For functions, the submit() returns the data.

For information on testing submits, refer to the Data Services Developer's Guide.

While Test View gives you a quick way to test simple update cases in the data services you create, for more substantial testing and troubleshooting you can use an update override class to inspect the decomposition mapping and update plan for the update.

The override class is also the mechanism you can use to extend and override the Mediator's default update processing. You can use it to implement updates for data services that would otherwise not support updates, such as non-relational sources. See Developing the UpdateOverride Class on page 9-6 for information about override classes.

Update Override Context

Although an update override class can programmatically access several update framework artifacts, including the update plan, decomposition map, and the tree of modified data objects, the content available at any time depends on the data service context, as follows:

Update Overrides and Physical Data Services

Considerations for implementing update override classes for physical level data services include the following:

For updated data objects bound to physical data services, further decomposition does not occur. Therefore, requesting a decomposition map or update plan in the override class of an object bound to such a service returns null.
If the data service is bound to a relational data source, returning true causes the Mediator to apply the changes currently indicated by the data object to the database. It does so using the optimistic locking strategy specified for the data service. (Note that if the data service is not bound to a relational data source, returning true will cause an exception.)
For physical data services, the update override can calculate a primary key value or perform other validations or calculations on the submitted data object. If an object bound to a physical data service is being updated in the context of an update to a higher-level data service object (that is, as a product of decomposition), changes in the physical update override (such as the primary key calculation) will be available when the higher-level update plan is applied. Therefore, if a primary key is calculated in the physical update override as part of a data object insert, the key will be available in the logical update plan, so that it can be assigned as a foreign key for the containing object.
A modified SDO that is passed to the physical level update override can see only those data object properties projected in the higher level data service. (See Figure 9-3.) To access the unprojected values as well, the update override must re-instantiate the data object.

Figure 9-3 Projected Data Objects

Projected Data Objects

Additional considerations concerning update overrides for relational data services include:

Update Override Programming Patterns

In an update override, you can modify the server-side update process as much or as little as you like, at any step of the way, to accomplish your goal. This section provides some code samples that illustrate common update override programming patterns, including:

Remember that an Update Override class is simply a Java class that implements the UpdateOverride interface. You can give the class any valid Java filename, but should use a meaningful name for common-sense reasons. After writing the class, you must register it with the data service, by setting the name of the class in the data service's Update Override Property field.

The class must include an implementation of the performChange( ) method; it is inside this method that you provide all custom code required for the programming task at hand. The performChange( ) method returns a boolean value that either continues or aborts processing by the Mediator, as discussed in How It Works: The Decomposition Process on page 2-18. The level of customization that you provide in your performChange( ) method determines whether you should return true or false, as noted in each of the sections below.

Overriding the Entire Decomposition and Update Process

To customize the entire decomposition and update process, the performChange( ) method can implement the following types of routines:

If your performChange( ) method does take over all processing, it should return false so that the Mediator does not proceed with automated decomposition.

Augmenting Data Object Content

The performChange( ) method can include code to inspect changed data object values and raise DataServiceException to signal errors, rolling back the transaction in such cases.

Return true to have the Mediator proceed with update propagation using the objects as changed.

Accessing the Data Service Mediator Context

To access the change plan and decomposition map for an update, you first must get the data service's Mediator context. The context enables you to view the decomposition map, produce an update plan, execute the update plan, and access the container data service instance for the data service object currently being processed.

Accessing the Decomposition Map

Once you have a decomposition map, you can use its toString() method to obtain the string rendering of the XML that map, as shown in Listing 9-5. (Note that although you can access the default decomposition map, you should not modify it.)

In addition to accessing the decomposition map, you can access the update plan in the override class. You can modify values in the tree, remove nodes, or rearrange them (to change the order in which they are applied). However, if you modify the update plan, you should execute the plan within the override if you want to keep the changes. As you modify the values in the tree, remove nodes or rearrange them, the update plan will track your changes automatically in the change list.

<xml-fragment xmlns:upd="update.dsmediator.ld.bea.com">
  <Binding>
    <DSName>ld:DataServices/CUSTOMERS.ds</DSName>
    <VarName>f1603</VarName>
  </Binding>
  <AttributeLineage>
    <ViewProperty>CUSTOMERID</ViewProperty>
    <SourceProperty>CUSTOMERID</SourceProperty>
    <VarName>f1603</VarName>
  </AttributeLineage>
  <AttributeLineage>
    <ViewProperty>CUSTOMERNAME</ViewProperty>
    <SourceProperty>CUSTOMERNAME</SourceProperty>
    <VarName>f1603</VarName>
  </AttributeLineage>
  <upd:DecompositionMap>
    <Binding>
      <DSName>ld:DataServices/getCustomerCreditRatingResponse.ds</DSName>
      <VarName>getCustomerCreditRating</VarName>
    </Binding>
    <AttributeLineage>
      <ViewProperty>CREDITSCORE</ViewProperty>
      <SourceProperty>
         getCustomerCreditRatingResult/TotalScore
      </SourceProperty>
      <VarName>getCustomerCreditRating</VarName>
    </AttributeLineage>
          ...

Customizing an Update Plan

After possibly validating or modifying the values in the submitted data object, the function retrieves the update plan by passing in the current data object to the following function:

After executing the update plan, the performChange( ) method should return false so that the Mediator does not attempt to apply the update plan.

The update plan lets you modify the values to be updated to the source. It also lets you modify the update order.

You can programmatically walk the update plan to view its contents by using your own method, similar to the navigateUpdatePlan(). As shown in Listing 9-6, navigateUpdatePlan() method takes a Collection object and uses an iterator to recursively walk the plan.

public boolean performChange(DataGraph datagraph){

	UpdatePlan up = 	DataServiceMediatorContext.currentContext().
				getCurrentUpdatePlan( datagraph );
	navigateUpdatePlan( up.getDataServiceList() );
	return true;
}

private void navigateUpdatePlan( Collection dsCollection ) {
	DataServiceToUpdate ds2u = null;
	for (Iterator it=dsCollection.iterator();it.hasNext();) {
            ds2u = (DataServiceToUpdate)it.next();
            
        // print the content of the SDO
	 System.out.println (ds2u.getDataGraph() );

	 // walk through contained SDO objects
        navigateUpdatePlan (ds2u.getContainedDSToUpdateList() );

Now consider an example in which a line item is deleted along with the order that contains it. Given the original data, Listing 9-7 illustrates an update plan in which item 1001 will be deleted from Order 100, and then the Order is deleted.

In this case, the execution of the update plan is as follows: before deleting the PO_CUSTOMERS, the contained SDOToUpdates routines are visited and processed. So the PO_ITEMS is deleted first and then PO_CUSTOMERS is deleted.

If the contents of the Update Plan are changed the new plan can then be executed. The update exit should then return false, signaling that no further automation should occur.

Executing an Update Plan

After modifying an update plan, you can execute it. Executing the update plan causes the Mediator to propagate changes to the indicated data sources.

Retrieving the Container of the Current Data Object

On a data service that is being processed for an update plan, you can get the container of the SDO being processed. The container must exist in the original changed object tree, as decomposed. If no container exists, null is returned. Consider the following example:

In this example, if in the update override class for the Orders data service the you ask to see the container, the Customer data service object for the Order instance being processed would be returned. If that Customer instance was in the update plan, then it would be returned. If it was not in the update plan, then it would be decomposed from CustOrders and returned.

The update plan only shows what has been changed. In some cases, the container will not be in the update plan. When the code asks for the container, it will be returned from the update plan if present; otherwise, it will be decomposed from the source SDO.

Invoking Other Data Service Functions and Procedures

Other data services may be accessed and updated from an update override. The data service the Mediator API can be used to access data objects, modify and submit them. Alternatively, the modified data objects can be added to the update plan and updated when the update plan is executed. If the data object is added to the update plan, it will be updated within the current context and its container will be accessible inside its data service update override.

If the DataService Mediator API is used to perform the update, a new DataService context is established for that submit, just as if it were being executed from the client. This submit() acts just like a client submit — changes are not reflected in the data object. Instead, the object must be re-fetched to see the changes made by the submit.

Capturing Runtime Data about Overrides in the Server Log

DSP uses the underlying WebLogic Server for logging. WebLogic logging is based on the JDK 1.4 logging APIs (available in the java.util.logging package). You can contribute to the log (from an update override) by acquiring a DataServiceMediatorContext instance, and then calling the getLogger() method on the context, as follows:

You can then contribute to the log by issuing the appropriate logger call with a specific log level. The log level implies the severity of the event. When WebLogic Server message catalogs and the NonCatalogLogger generate messages, they convert the message severity to a weblogic.logging.WLLevel object. A WLLevel object can specify any of the values listed in Table 9-4, from lowest to highest impact:

Table 9-4 WebLogic Server Log Level Definitions
Level	Description
DEBUG	Debug information, including execution times.
INFO	Normal events with informational value. This will allow you to see SQL that is executed against the underlying databases.
WARNING	Events that may cause errors.
ERROR	Events that cause errors.
NOTICE	Normal but significant events.
CRITICAL, ALERT, EMERGENCY	Significant events that require immediate intervention.

Given the specified logging level, the Mediator logs the information shown in Table 9-5.

Table 9-5 DSP Log Levels
Level	Information provided for...	Information captured
Notice or summary	Each submit from a client	Fully qualified data service name Invocation time Total execution time Invocation by user/group
Information or Detail	Each submit on a data service at any level	For a fully qualified data service name: Invocation time Number of times executed Total execution time For relational sources, per SQL statement type per table: SQL script Total execution time Number of times executed
Each update override invocation	Name of data service being overridden Number of times called Total execution time

<Nov 4, 2004 11:50:10 AM PST> <Notice> <LiquidData> <000000> <Demo - begin client sumbitted DS: ld:DataServices/Customer.ds>
<Nov 4, 2004 11:50:10 AM PST> <Notice> <LiquidData> <000000> <Demo - ld:DataServices/Customer.ds number of execution: 1 total execution time:171>
<Nov 4, 2004 11:50:10 AM PST> <Info> <LiquidData> <000000> <Demo - ld:DataServices/CUSTOMERS.ds number of execution: 1 total execution time:0>
<Nov 4, 2004 11:50:10 AM PST> <Info> <LiquidData> <000000> <Demo - EXECUTING SQL: update WEBLOGIC.CUSTOMERS set CUSTOMERNAME=? where CUSTOMERID=? AND CUSTOMERNAME=? number of execution: 1 total execution time:0>
<Nov 4, 2004 11:50:10 AM PST> <Info> <LiquidData> <000000> <Demo - ld:DataServices/PO_ITEMS.ds number of execution: 3 total execution time:121>
<Nov 4, 2004 11:50:10 AM PST> <Info> <LiquidData> <000000> <Demo - EXECUTING SQL: update WEBLOGIC.PO_ITEMS set ORDERID=? , QUANTITY=? where ITEMNUMBER=? AND ORDERID=? AND QUANTITY=? AND KEY=? number of execution: 3 total execution time:91>
<Nov 4, 2004 11:50:10 AM PST> <Notice> <LiquidData> <000000> <Demo - end clientsumbitted ds: ld:DataServices/Customer.ds Overall execution time: 381>

Default Optimistic Locking Policy: What it Means, How to Change

Locking mechanisms are used in numerous types of multi-user systems for concurrency control—to ensure that data is consistent, across transactions and regardless of the number of users acting on the system at the same time. Optimistic locking mechanisms are so-called because they typically only lock data at the time it is being updated (written to), not when it is being read.

DSP uses optimistic locking as its concurrency control policy, locking data only when updates are being attempted. When DSP receives submitted data graph, it compares the values of the data used to instantiate the original data objects with the original values in the data graph to ensure that the data was not changed by another user process during the time the data objects were being modified by a client application.

The Mediator compares fields from the original and the source; by default, Projected is used as the point of comparison (see Table 9-6).

You can specify the fields to be compared at the time of the update for each table. Note that primary key column must match, and BLOB and floating types might not be compared. Table 9-6 describes the options.

Table 9-6 Optimistic Locking Update Policy Options
Optimistic Locking Update Policy	Effect
Projected	Projected is the default setting. It uses a 1-to-1 mapping of elements in the SDO data graph to the data source to verify the "updateability" of the data source. This is the most complete means of verifying that an update can be completed, however if many elements are involved updates will take longer due to the greater number of fields to be verified.
Update	Only fields that have changed in your SDO data graph are used to verify the changed status of the data source.
Selected Fields	Selected fields are used to validate the changed status of the data source.

Client Application Developer's Guide