Please note that the specifications and other information contained herein are not final and are subject to change. The information is being made available to you solely for purpose of evaluation.
Indicates an error with the SyncProvider mechanism.
Package javax.sql.rowset.spi Description
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.)
The following classes and interfaces make up the javax.sql.rowset.spi
package:
SyncFactory
SyncProvider
SyncFactoryException
SyncProviderException
SyncResolver
XmlReader
XmlWriter
TransactionalWriter
The following interfaces, in the javax.sql package, are also part of the SPI:
RowSetReader
RowSetWriter
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
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.
RIOptimisticProvider
The default provider that the SyncFactory instance will
supply to a disconnected RowSet object when no provider
implementation is specified.
This synchronization provider uses an optimistic concurrency model,
assuming that there will be few conflicts among users
who are accessing the same data in a database. It avoids
using locks; rather, it checks to see if there is a conflict
before trying to synchronize the RowSet object and the
data source. If there is a conflict, it does nothing, meaning that
changes to the RowSet object are not persisted to the data
source.
RIXMLProvider
A synchronization provider that can be used with a
WebRowSet object, which is a rowset that can be written
in XML format or read from XML format. The
RIXMLProvider implementation does no checking at all for
conflicts and simply writes any updated data in the
WebRowSet object to the underlying data source.
WebRowSet objects use this provider when they are
dealing with XML data.
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.
Property Files - Properties specified in a standard property file.
This can be specified using a System Property or by modifying a standard
property file located in the platform run-time. The
reference implementation of this technology includes a standard property
file than can be edited to add additional SyncProvider objects.
JNDI Context - Available providers can be registered on a JNDI
context. The SyncFactory will attempt to load SyncProvider
objects bound to the context and register them with the factory. This
context must be supplied to the SyncFactory for the mechanism to
function correctly.
Details on how to specify the system properties or properties in a property file
and how to configure the JNDI Context are explained in detail in the
SyncFactory class description.
The SyncFactory generates a requested SyncProvider
object if the provider has been correctly registered. The
following policies are adhered to when either a disconnected RowSet object
is instantiated with a specified SyncProvider implementation or is
reconfigured at runtime with an alternative SyncProvider object.
If a SyncProvider object is specified and the SyncFactory
contains no reference to the provider, a SyncFactoryException is
thrown.
If a SyncProvider object is specified and the SyncFactory
contains a reference to the provider, the requested provider is supplied.
If no SyncProvider object is specified, the reference
implementation provider RIOptimisticProvider is supplied.
These policies are explored in more detail in the SyncFactory class.
Furthermore, the SyncProvider naming convention must be followed as
detailed in the SyncProvider class
description.
3.2 Grades
JSR 114 defines a set of grades to describe the quality of synchronization
a SyncProvider object can offer a disconnected RowSet
object. These grades are listed from the lowest quality of service to the highest.
GRADE_NONE - No synchronization with the originating data source is
provided. A SyncProvider implementation returning this grade will simply
attempt to write any data that has changed in the RowSet object to the
underlying data source, overwriting whatever is there. No attempt is made to compare
original values with current values to see if there is a conflict. The
RIXMLProvider is implemented with this grade.
GRADE_CHECK_MODIFIED_AT_COMMIT - A low grade of optimistic synchronization.
A SyncProvider implementation returning this grade
will check for conflicts in rows that have changed between the last synchronization
and the current synchronization under way. Any changes in the originating data source
that have been modified will not be reflected in the disconnected RowSet
object. If there are no conflicts, changes in the RowSet object will be
written to the data source. If there are conflicts, no changes are written.
The RIOptimisticProvider implementation uses this grade.
GRADE_CHECK_ALL_AT_COMMIT - A high grade of optimistic synchronization.
A SyncProvider implementation returning this grade
will check all rows, including rows that have not changed in the disconnected
RowSet object. In this way, any changes to rows in the underlying
data source will be reflected in the disconnected RowSet object
when the synchronization finishes successfully.
GRADE_LOCK_WHEN_MODIFIED - A pessimistic grade of synchronization.
SyncProvider implementations returning this grade will lock
the row in the originating data source that corresponds to the row being changed
in the RowSet object to reduce the possibility of other
processes modifying the same data in the data source.
GRADE_LOCK_WHEN_LOADED - A higher pessimistic synchronization grade.
A SyncProvider implementation returning this grade will lock
the entire view and/or table affected by the original query used to
populate a RowSet object.
3.3 Locks
JSR 114 defines a set of constants that specify whether any locks have been
placed on a RowSet object's underlying data source and, if so,
on which constructs the locks are placed. These locks will remain on the data
source while the RowSet object is disconnected from the data source.
These constants should be considered complementary to the
grade constants. The default setting for the majority of grade settings requires
that no data source locks remain when a RowSet object is disconnected
from its data source.
The grades GRADE_LOCK_WHEN_MODIFIED and
GRADE_LOCK_WHEN_LOADED allow a disconnected RowSet object
to have a fine-grained control over the degree of locking.
DATASOURCE_NO_LOCK - No locks remain on the originating data source.
This is the default lock setting for all SyncProvider implementations
unless otherwise directed by a RowSet object.
DATASOURCE_ROW_LOCK - A lock is placed on the rows that are touched by
the original SQL query used to populate the RowSet object.
DATASOURCE_TABLE_LOCK - A lock is placed on all tables that are touched
by the query that was used to populate the RowSet object.
DATASOURCE_DB_LOCK
A lock is placed on the entire data source that is used by the RowSet
object.
3.4 Updatable Views
A RowSet object may be populated with data from an SQL VIEW.
The following constants indicate whether a SyncProvider object can
update data in the table or tables from which the VIEW was derived.
UPDATABLE_VIEW_SYNC
Indicates that a SyncProvider implementation supports synchronization
to the table or tables from which the SQL VIEW used to populate a
a RowSet object is derived.
NONUPDATABLE_VIEW_SYNC
Indicates that a SyncProvider implementation does not support
synchronization to the table or tables from which the SQL VIEW
used to populate a RowSet object is derived.
3.5 Usage of SyncProvider Grading and Locking
In the example below, the reference CachedRowSetImpl implementation
reconfigures its current SyncProvider object by calling the
setSyncProvider method.
CachedRowSetImpl crs = new CachedRowSetImpl();
crs.setSyncProvider("com.foo.bar.HASyncProvider");
An application can retrieve the SyncProvider object currently in use
by a disconnected RowSet object. It can also retrieve the
grade of synchronization with which the provider was implemented and the degree of
locking currently in use. In addition, an application has the flexibility to set
the degree of locking to be used, which can increase the possibilities for successful
synchronization. These operation are shown in the following code fragment.
SyncProvider sync = crs.getSyncProvider();
switch (sync.getProviderGrade()) {
case: SyncProvider.GRADE_CHECK_ALL_AT_COMMIT
//A high grade of optimistic synchronization
break;
case: SyncProvider.GRADE_CHECK_MODIFIED_AT_COMMIT
//A low grade of optimistic synchronization
break;
case: SyncProvider.GRADE_LOCK_WHEN_LOADED
// A pessimistic synchronization grade
break;
case: SyncProvider.GRADE_LOCK_WHEN_MODIFIED
// A pessimistic synchronization grade
break;
case: SyncProvider.GRADE_NONE
// No synchronization with the originating data source provided
break;
}
switch (sync.getDataSourcLock() {
case: SyncProvider.DATASOURCE_DB_LOCK
// A lock is placed on the entire datasource that is used by the
// RowSet object
break;
case: SyncProvider.DATASOURCE_NO_LOCK
// No locks remain on the originating data source.
break;
case: SyncProvider.DATASOURCE_ROW_LOCK
// A lock is placed on the rows that are touched by the original
// SQL statement used to populate
// the RowSet object that is using the SyncProvider
break;
case: DATASOURCE_TABLE_LOCK
// A lock is placed on all tables that are touched by the original
// SQL statement used to populated
// the RowSet object that is using the SyncProvider
break;
It is also possible using the static utility method in the
SyncFactory class to determine the list of SyncProvider
implementations currently registered with the SyncFactory.
Enumeration e = SyncFactory.getRegisteredProviders();
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:
finding out whether the conflict involved an update, a delete, or an insert
getting the value in the data source that caused the conflict
setting the value that should be in the data source if it needs to be changed
or setting the value that should be in the RowSet object if it needs
to be changed
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.
Submit a bug or feature For further API reference and developer documentation, see Java SE Documentation. That documentation contains more detailed, developer-targeted descriptions, with conceptual overviews, definitions of terms, workarounds, and working code examples.