Re: XAResource implementation

Kris Jurka wrote:
>
>
> On Tue, 13 Nov 2007, Heikki Linnakangas wrote:
>
>> It looks like there is indeed a bug in the Postgres driver. I believe
>> it was introduced by the recent changes to keep the connection in
>> autocommit mode when no XA-transaction is in progress.
>>
>> It's this call sequence that borks it up:
>>
>> xares.start()
>> conn = xares.getConnection()
>> // do stuff
>> conn.close();
>> conn = xares.getConnection()
>> // do more stuff
>> xares.end()
>> xares.commit()
>>
>> The 2nd getConnection call inadvertently commits the transaction, and
>> opens another one.
>
> When I raised this issue, you said it can't/shouldn't happen.  What's
> different now?  It's still a bug in the driver, but is the calling code
> doing something it shouldn't?
>
> http://archives.postgresql.org/pgsql-jdbc/2007-06/msg00095.php

Oh, I had forgotten about that. The difference is the close() between
the getConnection()s.

I started having second doubts about that case you posted in June, so I
tested what Derby does if you call getConnection() twice, without a
close in between. You get an error ("Cannot close a connection while a
global transaction is still active."), so at least they consider it
incorrect as well.

Throwing an error in that case would be a nice thing for us to do as
well. Currently, we just silently rollback.

>> Patch attached. I also added a test for this in the test suite.
>
> transaction_timestamp() is an 8.2 feature, but since XA is an 8.1
> feature you should use now() instead.  (No need for a new patch, I've
> updated mine here.)

Thanks!

--
   Heikki Linnakangas
   EnterpriseDB   http://www.enterprisedb.com