The standard classes and interfaces that a third party vendor has to use in its implementation of a synchronization provider. These classes and interfaces are referred to as the Service Provider Interface (SPI). A vendor may have its implementation included on the JDBC web page that lists available SyncProvider implementations by sending email to jdbc@sun.com. Doing this helps make developers aware of the implementation. To make it possible for a RowSet object to use an implementation, the vendor must register it with the SyncFactory singleton. (See the class comment for SyncProvider for a full explanation of the registration process and the naming convention to be used.)

Table of Contents

1.0 Package Specification

The following classes and interfaces make up the javax.sql.rowset.spi package:

The following interfaces, in the javax.sql package, are also part of the SPI:

A SyncProvider implementation provides a disconnected RowSet object with the mechanisms for reading data into it and for writing data that has been modified in it back to the underlying data source. A reader, a RowSetReader or XMLReader object, reads data into a RowSet object when the CachedRowSet methods execute or populate are called. A writer, a RowSetWriter or XMLWriter object, writes changes back to the underlying data source when the CachedRowSet method acceptChanges is called.

The process of writing changes in a RowSet object to its data source is known as synchronization. The SyncProvider implementation that a RowSet object is using determines the level of synchronization that the RowSet object's writer uses. The various levels of synchronization are referred to as grades.

The lower grades of synchronization are known as optimistic concurrency levels because they optimistically assume that there will be no conflicts or very few conflicts. A conflict exists when the same data modified in the RowSet object has also been modified in the data source. Using the optimistic concurrency model means that if there is a conflict, modifications to either the data source or the RowSet object will be lost.

Higher grades of synchronization are called pessimistic because they assume that others will be accessing the data source and making modifications. These grades set varying levels of locks to increase the chances that no conflicts occur.

The lowest level of synchronization is simply writing any changes made to the RowSet object to its underlying data source. The writer does nothing to check for conflicts. If there is a conflict and the data source values are overwritten, the changes other parties have made by to the data source are lost.

The RIXMLProvider implementation uses the lowest level of synchronization and just writes RowSet changes to the data source. This is true because typically XML data sources do not enable transaction techniques for maintaining the integrity of data. However, specific standards groups have considered offering XML-based synchronization. For details, see 

     http://www.syncml.org

For the the next level up, the writer checks to see if there are any conflicts, and if there are, it does not write anything to the data source. The problem with this concurrency level is that if another party has modified the corresponding data in the data source since the RowSet object got its data, the changes made to the RowSet object are lost. The RIOptimisticProvider implementation uses this level of synchronization.

At higher levels of synchronization, referred to as pessimistic concurrency, the writer take steps to avoid conflicts by setting locks. Setting locks can vary from setting a lock on a single row to setting a lock on a table or the entire data source. The level of synchronization is therefore a tradeoff between the ability of users to access the data source concurrently and the ability of the writer to keep the data in the RowSet object and its data source synchronized.

It is a requirement that all disconnected RowSet objects (CachedRowSet, FilteredRowSet, JoinRowSet, and WebRowSet objects) obtain their SyncProvider objects from the SyncFactory mechanism.

The reference implementation (RI) provides two synchronization providers.

These SyncProvider implementations are bundled with the reference implementation, which makes them always available to RowSet implementations. SyncProvider implementations make themselves available by being registered with the SyncFactory singleton. When a RowSet object requests a provider, by specifying it in the constructor or as an argument to the CachedRowSet method setSyncProvider, the SyncFactory singleton checks to see if the requested provider has been registered with it. If it has, the SyncFactory creates an instance of it and passes it to the requesting RowSet object. If the SyncProvider implementation that is specified has not been registered, the SyncFactory singleton causes a SyncFactoryException object to be thrown. If no provider is specified, the SyncFactory singleton will create an instance of the default provider implementation, RIOptimisticProvider, and pass it to the requesting RowSet object.

If a WebRowSet object does not specify a provider in its constructor, the SyncFactory will give it an instance of RIOptimisticProvider. However, the constructor for WebRowSet is implemented to set the provider to the RIXMLProvider, which reads and writes a RowSet object in XML format.

See the SyncProvider class specification for further details.

Vendors may develop a SyncProvider implementation with any one of the possible levels of synchronization, thus giving RowSet objects a choice of synchronization mechanisms. A vendor can make its implementation available by registering the fully qualified class name with Oracle Corporation at jdbc@sun.com. This process is discussed in further detail below.

2.0 Service Provider Interface Architecture

  • 3.0 SyncProvider Implementer's Guide

    4.0 Resolving Synchronization Conflicts

    The interface SyncResolver provides a way for an application to decide manually what to do when a conflict occurs. When the CachedRowSet method acceptChanges finishes and has detected one or more conflicts, it throws a SyncProviderException object. An application can catch the exception and have it retrieve a SyncResolver object by calling the method SyncProviderException.getSyncResolver().

    A SyncResolver object, which is a special kind of CachedRowSet object or a JdbcRowSet object that has implemented the SyncResolver interface, examines the conflicts row by row. It is a duplicate of the RowSet object being synchronized except that it contains only the data from the data source this is causing a conflict. All of the other column values are set to null. To navigate from one conflict value to another, a SyncResolver object provides the methods nextConflict and previousConflict.

    The SyncResolver interface also provides methods for doing the following:

    When the CachedRowSet method acceptChanges is called, it delegates to the RowSet object's SyncProvider object. How the writer provided by that SyncProvider object is implemented determines what level (grade) of checking for conflicts will be done. After all checking for conflicts is completed and one or more conflicts has been found, the method acceptChanges throws a SyncProviderException object. The application can catch the exception and use it to obtain a SyncResolver object.

    The application can then use SyncResolver methods to get information about each conflict and decide what to do. If the application logic or the user decides that a value in the RowSet object should be the one to persist, the application or user can overwrite the data source value with it.

    The comment for the SyncResolver interface has more detail.

    5.0 Related Specifications

    6.0 Related Documentation