openjpa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Pinaki Poddar" <ppod...@bea.com>
Subject RE: [jira] Updated: (OPENJPA-464) Performance improvement with Statement Batching support
Date Tue, 18 Dec 2007 19:30:06 GMT
Hi Teresa,
 > Our internal design was in the extension of the OperationUpdateManager 
 Will you please clarify what is meant by 'Our internal design'?  

 > I did this before we can't change the code in the DBDictionary at the first place.
 Not clear what you mean. The patch does modify DBDictionary.

 UpdateManager interface is used to customize the way that updates are made to database records.

 UpdateManager works in conjunction with PreparedStatementManager which manages prepared statement
execution. 
 So as you are supporting batching of SQL, a extended PreparedStatementManager is the correct
way to go which is configured via UpdateManager.
 Whereas, your patch modifies PreparedStatementManagerImpl to accomodate batching logic. I
do not think this is in-line with original designer's view. Otherwise AbstractUpdateManager
will not have a newPreparedStatementManager() method.


> Therefore, there is no overhead for the non-batch process - the implementation is separated
in a way that you don't need to use the plugin class.

I am not making any efficiency/performance argument against your patch. I am making a design
argument. And I am in favor of a plug-in UpdateManager that supports batching alongside the
existing one that does not. That is in-line with plug-in architecture of OpenJPA for feature
extension.  



-----Original Message-----
From: Teresa Kan [mailto:tckan1@gmail.com] 
Sent: Tuesday, December 18, 2007 11:06 AM
To: dev@openjpa.apache.org
Subject: Re: [jira] Updated: (OPENJPA-464) Performance improvement with Statement Batching
support

Hi Pinaki,
Our internal design was in the extension of the OperationUpdateManager, so the batch and non-batch
support was separated. I did this before we can't change the code in the DBDictionary at the
first place. When I moved the function from our internal implementation to the openjpa layer,
I think to enable the batch from DBDictionary is a better decision. From usability point of
view, it is better to configure the batchLimit under the DBDictionary instead of UpdateManager.
My implementation tested the batchLimit at the constructor of PreparedStatementManagerImpl
, if the Dictionary does not support batch, then it will be disabled and the non-batch process
is used instead. The MAP in this class will not be initiatized until it is necessary to use.
Therefore, there is no overhead for the non-batch process - the implementation is separated
in a way that you don't need to use the plugin class. By default, most of DB Dictionaries
are disabled with the exception of DB2 and Oracle.

Teresa

On Dec 18, 2007 12:36 PM, Pinaki Poddar <ppoddar@bea.com> wrote:

> Hi Teresa,
>
>  Batching support should be added to a separate implementation of 
> PreparedStatementManager rather than existing 
> PreparedStatementManagerImpl trying to act on both batch and non-batch 
> mode. A new BatchPreparedStatementManager can be plugged-in by 
> overwriting AbstractUpdateManager.newPreparedStatementManager().
>
> That way, one will not require a Map in 
> BatchingPreparedStatementManager but only a instance variable e.g. 'PreparedStatement
_sql'.
> The flushInternal() checks whether the current SQL requires a new 
> PreparedStatement or the existing one will do. If it requires a new 
> PreparedStatement then existing _sql is executed + _sql is reassigned 
> the current one. Otherwise, _sql.addBatch(current).
>
> This design will keep batch/non-batch support separate and will help 
> maintenance in future.
>
>
>
>
> -----Original Message-----
> From: Teresa Kan [mailto:tckan1@gmail.com]
> Sent: Tuesday, December 18, 2007 9:30 AM
> To: dev@openjpa.apache.org
> Subject: Re: [jira] Updated: (OPENJPA-464) Performance improvement 
> with Statement Batching support
>
> I would like to hear any comment from the community and would like to 
> commit the changes before the holidays. If there is no objection by 
> end of tomorrow. The patch will be committed.
>
> Thanks and happy holiday!
> Teresa
>
> On Dec 13, 2007 4:15 PM, Teresa Kan (JIRA) <jira@apache.org> wrote:
>
> >
> >     [
> > https://issues.apache.org/jira/browse/OPENJPA-464?page=com.atlassian
> > .j ira.plugin.system.issuetabpanels:all-tabpanel]
> >
> > Teresa Kan updated OPENJPA-464:
> > -------------------------------
> >
> >    Attachment: OPENJPA-464.patch
> >
> > Attach the patch including the OPENJPA documentation changes.
> >
> > > Performance improvement with Statement Batching support
> > > -------------------------------------------------------
> > >
> > >                 Key: OPENJPA-464
> > >                 URL: https://issues.apache.org/jira/browse/OPENJPA-464
> > >             Project: OpenJPA
> > >          Issue Type: Bug
> > >          Components: kernel
> > >    Affects Versions: 1.1.0
> > >            Reporter: Teresa Kan
> > >             Fix For: 1.1.0
> > >
> > >         Attachments: OPENJPA-464.patch, statement batch 
> > > design_1211.doc
> > >
> > >
> > > The current OpenJPA implementation did not provide the SQL 
> > > statement
> > batching. All SQL statements will be executed one statement at a 
> > time to the database. Consequently, the runtime performance was 
> > decreased due to lots of database flows. JDBC Specification provides 
> > the batch capability for insert, update and delete statements 
> > through the
> addBatch() and executeBatch() APIs.
> > We should be able to take advantage of this capability to support 
> > SQL statement batching in OpenJPA.
> > > According to the old version of the OpenJPA manual (i.e., Kodo),
> > statement batching was part of the initial functions. Conscious 
> > decision by BEA that this function was not contributed back to 
> > OpenJPA. We can still use this info as the implementation base with 
> > some
> modifications.
> > > I have completed the work for this statement batching support and 
> > > the
> > patch has been tested by CTS against Derby and DB2, OPENJPA 
> > regression test as well as our internal FVT test bucket.  The 
> > following section describes the design and implementation info. I 
> > also attached the whole design documentation and the patch in this 
> > jira. Once the design and implementation are accepted, then I will 
> > update the OPENJPA manual to include this function. Thanks,
> > > Design and implementation:
> > > *     Configuration:
> > > o     Batch Limit value:
> > > §     0 - Disable batch support.
> > > §     -1 - Unlimited number of statements for a batch.
> > > §     Any positive number - Maximum number of statements for a batch.
> > > o     By default, the batch support is based on each Dictionary to
> > define the default batch limit. Currently only DB2 and Oracle 
> > dictionaries are set the default batch limit to 100. The default 
> > batch limit for rest of the dictionaries is set to zero (disabled).
> > > o     To enable the batch support, user can specify the following
> > property in the persistence.xml file:
> > > <property name="openjpa.jdbc.DBDictionary" value="BatchLimit=25"/> 
> > > or <property name="openjpa.jdbc.DBDictionary"
> > > value="db2(batchLimit=25)"/>
> > > *     Basic design is to cache all the insert/update/delete statements
> > during the execution of the
> > PreparedStatementManagerImpl.flushInternal()
> > method. There is a cache structure which uses the LinkHashMap to 
> > maintain the order of the SQL statements for execution:
> > > o     _cacheSql - a LinkHashMap to store the rows that associate with
> > one PrepareStatement. Key: SQL statement string; Value: array list 
> > of
> rows.
> > > During the PreparedStatementManagerImpl.flush() process, it will 
> > > go
> > through the cache to prepare the SQL statement; add the statement to 
> > the batch; and execute the batch when the batch limit is reached or 
> > all the rows are processed for that statement. Validate the update 
> > count after the
> > executeBatch() method.
> > > *     If the batch limit =0 (disabled), execute the statement as the
> > normal process; no need to use the batching process.  Same rule 
> > applies to the statement that only has one row, execute it as the 
> > normal
> process.
> > > *     The batch process will be disabled if the primary key generation
> > is used the Identity strategy. When the GeneratedType=IDENTITY, we 
> > need to get the ID value right away for the in-memory entity to use.
> > Therefore, we can't batch this kind of statement.
> > > *     Batch exception process: a checkUpdateCount() is used to
> validate
> > the batch process after the executeBatch(). According to the 
> > javadoc, there are three cases to consider:
> > > o     Case of EXECUTE_FAILED: (-3):
> > > §     This is a failure case. If the action is UPDATE or there is
> > FailedObject, treats it as OptimisticException. Otherwise, throws 
> > the SQLException.
> > > §     This is the same process as current implementation.
> > > o     Case of  SUCCESS_NO_INFO: (-2):
> > > §     We treat this as successful case and log the info in the log.
> > > o     Case of 0:
> > > §     If there is a FailedObject or the action is INSERT, then throws
> > the SQLException. Otherwise, treats it as successful case.
> >
> > --
> > This message is automatically generated by JIRA.
> > -
> > You can reply to this email to add a comment to the issue online.
> >
> >
>
> Notice:  This email message, together with any attachments, may 
> contain information  of  BEA Systems,  Inc.,  its subsidiaries  and  
> affiliated entities,  that may be confidential,  proprietary,  
> copyrighted  and/or legally privileged, and is intended solely for the 
> use of the individual or entity named in this message. If you are not 
> the intended recipient, and have received this message in error, 
> please immediately return this by email and then delete it.
>

Notice:  This email message, together with any attachments, may contain information  of  BEA
Systems,  Inc.,  its subsidiaries  and  affiliated entities,  that may be confidential,  proprietary,
 copyrighted  and/or legally privileged, and is intended solely for the use of the individual
or entity named in this message. If you are not the intended recipient, and have received
this message in error, please immediately return this by email and then delete it.

Mime
View raw message