[hibernate-issues] [Hibernate-JIRA] Commented: (HHH-3360) Custom Oracle Batcher to allow batch updates for versioned data

[Manuel Dominguez Sarmiento (JIRA)]

    [ http://opensource.atlassian.com/projects/hibernate/browse/HHH-3360?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=44340#comment-44340 ] 

Manuel Dominguez Sarmiento commented on HHH-3360:
-------------------------------------------------

Please add it to the wiki, at least. We've put significant work into this and would appreciate this being made more visible for fellow Oracle users.

Some observations:

1) Cannot these classes be packaged in a companion hibernate-oracle.jar ? I understand that you do not generally rely on driver-specific implementation classes, but given that Oracle is used by so many people, I guess it's worth it. The separate packaging approach is used by many frameworks (e.g. Quartz)

2) We tried to migrate our apps to Hibernate 4.x but found that there is no equivalent hook in order to implement a custom batching strategy. Is this being addressed?

3) This should probably be filed as a separate issue, but while debugging this custom batching implementation, we noticed that internal state should be reset/cleared/cleaned up on closeStatements() as well as in abortBatch(). Our latest implementation takes this into account. However, BatchingBatcher (which is the default everyone uses) does not clean up its internal state in these cases, which may cause strange problems in some edge cases. The following fields are at risk:

	private int batchSize;
	private Expectation[] expectations;



> Custom Oracle Batcher to allow batch updates for versioned data
> ---------------------------------------------------------------
>
>                 Key: HHH-3360
>                 URL: http://opensource.atlassian.com/projects/hibernate/browse/HHH-3360
>             Project: Hibernate Core
>          Issue Type: Improvement
>          Components: core
>    Affects Versions: 3.2.0.ga, 3.2.1, 3.2.2, 3.2.3, 3.2.4, 3.2.4.sp1, 3.2.5, 3.2.6, 3.3.0.CR1
>         Environment: Oracle 10g R1, 10g R2, 11g R1 (have not tried previous Oracle versions), 11g R1 drivers (older drivers should also work)
>            Reporter: Manuel Dominguez Sarmiento
>            Assignee: Steve Ebersole
>            Priority: Minor
>         Attachments: C3P0OracleBatchingBatcherFactory.java, C3P0OracleBatchingBatcher.java, OracleBatchingBatcherFactory.java, OracleBatchingBatcher.java, OracleBatchingBatcher.java, OracleBatchingBatcher.java
>
>   Original Estimate: 0.5h
>  Remaining Estimate: 0.5h
>
> We have developed a custom Oracle Batcher which allows batching versioned data. The Oracle JDBC driver does not return update counts when using the standard JDBC 2.0 batching mechanism, however the proprietary Oracle batching mechanism allows obtaining the total batch row update count. The update counts are absolutely necessary to detect stale updates.
> Although it is not exactly the same, the total row update count is actually enough information to be able to batch versioned data and still detect stale updates.
> We'd like to contribute the attached files. They have a compile time dependency on Oracle JDBC. If this is not acceptable, it could be easily solved by using reflection.
> Another Batcher is provided for when the Oracle connection is being managed through c3p0 (a common deployment scenario). This has a compile time dependency on c3p0.
> A few "dirty" tricks were necessary to pull this off without patching other classes. Specifically, it was necessary to override Java private semantics to obtain BasicExpectation.expectedRowCount. This could be easily solved by adding an accessor method to the Expectation interface.
> There is one issue which we are not completely sure of, however so far we have not found any problems. When the Expectation is NONE, there is no way to check whether the total row count is correct or not, even if other batched updates do have expectations with expected row counts. Our understanding is that actually, since batching requires all statements to be of the same type (since the same PreparedStatement / CallableStatement is being used), then either ALL expectations will be NONE, or all will have an expected row count. We'd welcome comments from the Hibernate team. This could also be probably handled better by improving the Expectation interface.
> Oracle JDBC docs that explain the Oracle batching model: http://download.oracle.com/docs/cd/B28359_01/java.111/b31224/oraperf.htm#i1059054
> As expected, implementing this solution has resulted in drastical improvement in batch processing.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira