Thread: XADataSource interface
Hello, I posted an updated version of the XADataSource implementation few weeks ago, but haven't received any comments: http://archives.postgresql.org/pgsql-jdbc/2005-09/msg00131.php Has anybody looked at it yet? - Heikki
Heikki, I will look at this shortly, is this a diff from a relatively current HEAD ? Dave On 25-Oct-05, at 1:01 PM, Heikki Linnakangas wrote: > Hello, > > I posted an updated version of the XADataSource implementation few > weeks ago, but haven't received any comments: > > http://archives.postgresql.org/pgsql-jdbc/2005-09/msg00131.php > > Has anybody looked at it yet? > > - Heikki > > ---------------------------(end of > broadcast)--------------------------- > TIP 4: Have you searched our list archives? > > http://archives.postgresql.org > >
Thanks. Yes, it's a diff from HEAD. Just checked, it still applies cleanly. It only modifies build.xml, everything else is in new files, so it shouldn't bit rot easily anyway. On Wed, 26 Oct 2005, Dave Cramer wrote: > Heikki, > > I will look at this shortly, is this a diff from a relatively current HEAD ? > > Dave > On 25-Oct-05, at 1:01 PM, Heikki Linnakangas wrote: > >> Hello, >> >> I posted an updated version of the XADataSource implementation few weeks >> ago, but haven't received any comments: >> >> http://archives.postgresql.org/pgsql-jdbc/2005-09/msg00131.php >> >> Has anybody looked at it yet? >> >> - Heikki >> >> ---------------------------(end of broadcast)--------------------------- >> TIP 4: Have you searched our list archives? >> >> http://archives.postgresql.org >> >> > > - Heikki
On Tue, 25 Oct 2005, Heikki Linnakangas wrote: > I posted an updated version of the XADataSource implementation few weeks ago, > but haven't received any comments: > Dave has committed this and I've made some further integration fixes to it, but some of the tests you've provided fail with the attached errors. I don't know much about XA, so I wanted to run some potential fixes by you guys first. Thoughts? Kris Jurka
Attachment
On Sat, 29 Oct 2005, Kris Jurka wrote:
> On Tue, 25 Oct 2005, Heikki Linnakangas wrote:
>
>> I posted an updated version of the XADataSource implementation few weeks
>> ago, but haven't received any comments:
>>
>
> Dave has committed this and I've made some further integration fixes to it,
> but some of the tests you've provided fail with the attached errors. I don't
> know much about XA, so I wanted to run some potential fixes by you guys
> first. Thoughts?
It seems I forgot to run the tests myself before submitting, sorry about
that...
The test failure at testOnePhase was caused by the inverted equals test
you already noticed and fixed in your patch.
The failures at testRecover and testRollback were because
PGXAConnection.rollback method called Xid.equals with null argument.
I added a null-check for that, see attached patch.
I didn't understand this part of your patch:
--------------------------------------------------------------------------
diff -c -r1.2 PGXAConnection.java
*** org/postgresql/xa/PGXAConnection.java 29 Oct 2005 18:59:39 -0000
1.2
--- org/postgresql/xa/PGXAConnection.java 29 Oct 2005 19:05:09 -0000
***************
*** 172,178 ****
throw new PGXAException(GT.tr("Prepare called before end"),
XAException.XAER_INVAL);
state = STATE_IDLE;
! currentXid = null;
if (!conn.haveMinimumServerVersion("8.1"))
throw new PGXAException(GT.tr("Server versions prior to 8.1
do not support two-phase
commit."), XAException.XAER_RMERR);
--- 172,178 ----
throw new PGXAException(GT.tr("Prepare called before end"),
XAException.XAER_INVAL);
state = STATE_IDLE;
! // currentXid = null;
if (!conn.haveMinimumServerVersion("8.1"))
throw new PGXAException(GT.tr("Server versions prior to 8.1
do not support two-phase
commit."), XAException.XAER_RMERR);
--------------------------------------------------------------------------
Why is that? It looks wrong to me.
- Heikki
Attachment
On Sun, 30 Oct 2005, Chris Smith wrote: > Heikki Linnakangas wrote: >> The failures at testRecover and testRollback were because >> PGXAConnection.rollback method called Xid.equals with null argument. >> I added a null-check for that, see attached patch. > > I didn't look too closely at the patch... but the API docs for Object.equals > say clearly that this shouldn't be a problem. For any non-null reference x, > x.equals(null) should evaluate to false. I suspect that the correct solution > to this problem is to fix the implementation of equals, and not to add a > check against null where you did. True. However, since the object is supplied by the transaction manager implementation, which we have no control over, I'd say it's better to not make any assumptions about it that we don't have to. - Heikki
On Sun, 30 Oct 2005, Heikki Linnakangas wrote: > The failures at testRecover and testRollback were because > PGXAConnection.rollback method called Xid.equals with null argument. > I added a null-check for that, see attached patch. Applied. I'll put out a new dev release with the xa patches later tonight. Can you clarify what does/doesn't work in xa for the docs? > I didn't understand this part of your patch: > Well, that's why I emailed you guys. I was initially thinking that the currentXid shouldn't be disassociated from the xaconnection until a new transaction was actually started and it could simply commit/rollback the underlying connection if a new one wasn't started. I see that's not the case now. Kris Jurka
On Mon, 31 Oct 2005, Kris Jurka wrote: > > > On Sun, 30 Oct 2005, Heikki Linnakangas wrote: > >> The failures at testRecover and testRollback were because >> PGXAConnection.rollback method called Xid.equals with null argument. >> I added a null-check for that, see attached patch. > > Applied. I'll put out a new dev release with the xa patches later tonight. > Can you clarify what does/doesn't work in xa for the docs? Thanks! English is not my native language, so please check spelling: " The JTA specification requires an XADataSource implementation to support transaction interleaving (see JTA spec section 3.4.4). However, since PostgreSQL server doesn't support it natively, the JDBC driver cannot support it either. The same goes for suspend/resume. In practice, transaction managers don't need transaction interleaving or suspend/resume to function correctly. Some commercial databases also don't implement them natively, but emulate them in the JDBC driver. That's error prone; PostgreSQL driver will throw an XAException instead. " The docs should also include instructions for using the XADataSource. Example configurations for various application servers would be a nice too. - Heikki