geronimo-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Lin Sun" <linsun....@gmail.com>
Subject RE: daytrader with Tranql Oracle XA resource adapter
Date Mon, 17 Jul 2006 16:13:11 GMT
Hi,

Sorry to get back to you late... got busy on other stuff.

Yes with the generic tranql wrapper, setting the commitBeforeAutocommit to
true would force the commit happen right away.

However this commitBeforeAutocommit isn't available for tranql XA oracle
wrapper.  For Oracle wrapper's LocalMCF, I can just change:

super(new OracleDataSource(), new OracleExceptionSorter(), false);

to

super(new OracleDataSource(), new OracleExceptionSorter(), true);

However, for Oracle wrapper's XAMCF class, the parent class
AbstractXADataSourceMCF's contructor only takes the first 2 parms. Here are
two options I can think of:  

1st option: add the "boolean commitBeforeAutocommit" parm into the
AbstractXADataSourceMCF's contructor, which will also require changes to the
localTransactionCommit method in ManagedXAConnection.   Also all the other
vendor wrappers (db2, derby, mysql) will have to be updated.

2nd option: create the OracleManagedXAConnection class that extends
AbstractManagedConnection.

I would vote for 1st option as we have a quite big number of buggy database
vendors.   What do you think?

Lin

-----Original Message-----
From: David Jencks [mailto:david_jencks@yahoo.com] 
Sent: Friday, July 07, 2006 12:39 PM
To: user@geronimo.apache.org
Subject: Re: daytrader with Tranql Oracle XA resource adapter

Can you (or have you already) try using the generic tranql wrapper  
with the oracle Driver class and set the commitBeforeAutocommit  
property to true?  This definitely _ought_ to make the commit happen  
immediately: if it does, the oracle driver has a bug and we need to  
modify the oracle wrapper to work around it.

thanks
david jencks

On Jul 7, 2006, at 8:25 AM, Lin Sun wrote:

> Hi there,
>
> Thanks for the pointer!  IIUC, if the application uses  
> connection.commit,
> tranql will translate that as setautocommit(true) when the
> commitBeforeAutoCommit is set to false.   I believe this is what I  
> have in
> my environment.
>
> However, I still don't understand why it makes such a big  
> difference for
> Oracle.  I have loaded daytrader on another J2EE application server
> (WebSphere Application Server) and found out the same order  
> committed to the
> oracle database right away.  With that, I would not categorize it  
> as one of
> the problematic drivers you mentioned before.
>
> Since derby and db2 XA adapter don't have this "delayed committing"  
> problem
> at all, I compared the two Tranql vendors' code with the Oracle  
> vendor code.
> Basically the vendor code contains the vendor specific MCF class  
> that is
> created based on the datasource from the vendor (for example,
> OracleXADataSource).  The MCF also sets/gets the datasource's  
> different
> properties (like databasename, username, password, servername,  
> etc.).  I
> still don't get much clue as to why Oracle is special with Tranql.
>
> I think what I'll do is to come up with a simple application that  
> can be
> loaded into Eclipse and step into the connection.commit() code from  
> Eclipse.
> Too bad that I haven't figured out how to get daytrader running from
> Eclipse.:-(
>
> Time to study the new plugin Sachin puts out.:-)
>
> Lin
>
>
>
>
>
> -----Original Message-----
> From: Mario Rübsam [mailto:mr@coderesearch.com]
> Sent: Thursday, July 06, 2006 2:37 PM
> To: user@geronimo.apache.org
> Subject: Re: daytrader with Tranql Oracle XA resource adapter
>
> Lin Sun wrote:
>> Hi there,
>>
>> Do you know how to log the commits on the oracle side?  I 'll be  
>> happy to
>> check that out if I know how to do it.
>
> I don't have a clue about the Oracle tools. But there must be some  
> admin
> tool to show the stats of the server, the connections, threads, ...
> There is normally a transaction counter in these tools. Another way  
> to find
> it out is to look into the driver itself and figure out what is  
> going on
> in the setAutoCommit() method. But this is not legal in every  
> country for
> closed source interfaces.
>
>>
>> I posted another note yesterday.   The application was not using
> AutoCommit.
>> Instead connection.commit() is being used.
>
> That is correct, but to commit the local transaction TranQL use the
> setAutoCommit(true); method, so if the JDBC driver is spec  
> compliant the
> transaction must be commited ... so far the theory, in the real world
> some of the drivers don't do it for performance or whatever reasons.
>
> There was already a discussion about this problem here:
> http://www.mail-archive.com/user@geronimo.apache.org/msg03450.html
>
> Thanks,
> Mario
>
>
>
>>
>> Lin
>>
>> -----Original Message-----
>> From: Mario Rübsam [mailto:mr@coderesearch.com]
>> Sent: Wednesday, July 05, 2006 4:59 PM
>> To: user@geronimo.apache.org
>> Subject: Re: daytrader with Tranql Oracle XA resource adapter
>>
>> I don't have experience with the Oracle Adapter. Can you log the  
>> commits
> on
>> the Oracle side to see if any commits come through? I had a similar
> problem
>> with MaxDB and it seems that the transactions are not closed  
>> because the
>> JDBC driver for MaxDB did no commit on the connection if the  
>> AutoCommit
> was
>> switched from false to true. The JDBC spec says if the value  
>> changed the
>> connection must be commited but the driver does not. Thats why I use
>> CommitBeforeAutCommit now.
>> Can you try to use the generic adapter to connect to Oracle and  
>> see if it
>> happens also?
>>
>> Thanks,
>> Mario
>>
>>
>> Lin Sun wrote:
>>> Hi there,
>>>
>>> Thanks - I remember seeing your posting on CommitBeforeAutCommit  
>>> setting
>> in
>>> the generic adapter.    The CommitBeforeAutCommit setting is not  
>>> avail in
>>> the Oracle XA resource adapter.
>>>
>>> Lin
>>>
>>> -----Original Message-----
>>> From: Mario Rübsam [mailto:mr@coderesearch.com]
>>> Sent: Thursday, June 29, 2006 4:56 PM
>>> To: user@geronimo.apache.org
>>> Subject: Re: daytrader with Tranql Oracle XA resource adapter
>>>
>>> Hi,
>>>
>>> the CommitBeforeAutCommit setting in the generic adapter call
>>> a commit on the local connection before it calls setAutoCommit 
>>> (true).
>>> Some JDBC drivers act not like specified in JDBC and don't call
>>> a commit when the AutoCommit value changes with setAutoCommit(true)
>>> so CommitBeforeAutCommit is a workaround to get the transaction
>>> commited on these drivers (MySQL, MaxDB, PostgreSQL in some cases).
>>> I don't know about the XA driver and Oracle but it could be a  
>>> similar
>>> problem.
>>>
>>>
>>> -Mario
>>>
>>>
>>>
>>> Lin Sun wrote:
>>>> Hi there,
>>>>
>>>> I have been playing with the daytrader with Tranql Oracle XA  
>>>> resource
>>>> adapter (tranql-connector-oracle-xa-1.1.rar).  After I tweaked the
>>>> plan a bit and I am able to get daytrader running well with oracle!
>>>>
>>>> One thing I noticed is that if I perform any trading activities  
>>>> as a
>>>> user (say uid:1, account_id 50), I get the message that my order  
>>>> has
>>>> been completed almost immediately.  However, the order isn't  
>>>> committed
>>>> into the oracle database until quite a while.  I don't know the  
>>>> exact
>>>> timing but seems to be over 20-30 minutes.
>>>>
>>>> Here is what I did to check if the order is recorded in the  
>>>> database:
>>>>
>>>> Select * from orderejb where account_id=50;
>>>>
>>>> If the order is committed into the database, I would expect a  
>>>> row that
>>>> represents my previous order to be returned as the output of the  
>>>> sql
>>>> command.
>>>>
>>>> If I run the same scenario with db2 or derby, I can see my order
>>>> recorded in the orderejb table immediately.
>>>>
>>>> I looked at the tranql oracle vendor code but I didn't see anywhere
>>>> specifying the commit timing thing.  The generic resource  
>>>> adapter does
>>>> offer an param called CommitBeforeAutCommit and by setting it to  
>>>> true
>>>> would force the commit to happen immediately.   Can someone shed  
>>>> some
>>>> light on this?
>>>>
>>>> Thanks, Lin
>>>>
>>>
>>
>>
>


Mime
View raw message