Scott,
For the majority of users having a query execute outside of the transaction will provide efficiency benefits and will have no impact on behaviour. There are fringe cases where this will be problematic but as long as we provide these fringe users with a straightforward mechanism for forcing the queries into the transaction then we have struck a balance between technical and practical functionality. I am not talking about all queries here and I am not talking about queries issued after the user has initiated changes in the EM just those queries that can be expected to successfully complete for the majority of situations.
I do think the NativeQueries should by default be issued in the transaction as these are already special cases and we cannot easily determine the user's intentions.
--Gordon
-----Original Message-----
From: Scott.Oaks_at_Sun.COM [mailto:Scott.Oaks_at_Sun.COM]On Behalf Of Scott
Oaks
Sent: Friday, September 01, 2006 3:24 PM
To: persistence_at_glassfish.dev.java.net
Cc: Binod P G
Subject: RE: [Issue 1059] New - native queries not being committed
On Fri, 2006-09-01 at 14:16, Gordon Yorke wrote:
> Having a query use the transactional connection in TopLink is not as simple as
> it seems and caries with it some side effects on subsequent queries. I think
> we should have an option to force the query into the transaction that can be
> initiated by the user and we should attempt to detect pessimistic lock clauses.
It strikes me that this is antithetical to the entire purpose of having
a container with CMT. As a developer, I shouldn't need to initiate
anything wrt transaction boundaries.
Aren't there also other things that could cause an update within the
transaction (like calling a statement with a trigger, or a stored
procedure)? It seems like it's impossible to know for an arbitrary set
of calls within a transaction whether the database has modified the data
the select should see -- so executing the statement outside of the
expected transaction seems dangerous to me in any case.
-Scott
> --Gordon
>
> -----Original Message-----
> From: Marina.Vatkina_at_Sun.COM [mailto:Marina.Vatkina_at_Sun.COM]On Behalf Of
> Marina Vatkina
> Sent: Friday, September 01, 2006 1:37 PM
> To: persistence_at_glassfish.dev.java.net
> Cc: Binod P G
> Subject: Re: [Issue 1059] New - native queries not being committed
>
>
> Hi Mitesh,
>
> I don't think that the answer is in the connection pool code. Non-tx connections
> are not suitable for select ... for update, as such selects must be transactional.
>
> Gordon,
>
> Unless there is a complete parser in place that can understand what exactly a
> native query is doing, I would think that native queries should be excluded
> from non-tx connections optimizations.
>
> thanks,
> -marina
>
> Mitesh Meswani wrote:
> >
> > Gordon Yorke wrote:
> >
> >>Hello Mitesh,
> >> In July (around the 20th) code was changed where if the SELECT was executed before a flush() or bulk update in the transaction then it would use the non-tx connection. (Select for update was not considered).
> >>
> >> Is it possible that the connection pool was giving us the tx connection anyway?
> >>
> > We will have to wait for Binod's reply. But, looking at the fisheye logs
> > it does not seem that there was any change in the jdbcra module either
> > in trunk <http://fisheye5.cenqua.com/changelog/glassfish/jdbcra> or in
> > the branch
> > <http://fisheye5.cenqua.com/changelog/%7Ebr=SJSAS90_FCS_BRANCH/glassfish/jdbcra>
> > post Marrch 2006 so most probably the code change above should be the
> > root cause
> >
> >>A bug should be entered for the custom SQL SELECT FOR UPDATE case.
> >>
> >>
> > Issue 1059 corresponds to this issue. Any hints on how it can be resolved?
> >
> > Thanks,
> > Mitesh
> >
> >>--Gordon
> >>
> >>-----Original Message-----
> >>From: Mitesh.Meswani_at_Sun.COM [mailto:Mitesh.Meswani_at_Sun.COM]On Behalf Of
> >>Mitesh Meswani
> >>Sent: Friday, September 01, 2006 3:10 AM
> >>To: persistence
> >>Cc: Binod P G
> >>Subject: Re: [Issue 1059] New - native queries not being committed
> >>
> >>
> >>Hi Tom, Gordon,
> >>
> >>I am trying to investigate this issue. I observed that current toplink
> >>code uses non-tx connection to execute the query. To experiment I
> >>changed toplink code to use the transactional connection
> >>(maindataSource) instead of non-tx connection (readDatasource) to
> >>execute the query and the hang goes away.
> >>Looks like the cause for this issue is one of following.
> >>(1). Toplink was changed to use non-tx connection between June 29th
> >>(date of merge that corresponds to build 8) to Aug 30 (date of merge
> >>that corresponds to build 9). Do you think it is the case? Or any other
> >>changes that you might suspect?
> >>(2). The non-tx connection pool impl of glassfish has changed and is not
> >>properly delisting the connection.
> >>
> >>Hi Binod,
> >>Stepping throu gh toplink's code, I observed that toplink does properly
> >>close the non-tx connection after executing the query. Do you think
> >>anything has changed in the connection pool impl in this area between
> >>UR1 build 8 and 9 that correposnds to (2) above ? Any other hints you
> >>can think of to debug this issue?
> >>
> >>Thanks,
> >>Mitesh
> >>
> >>sdo_at_dev.java.net wrote:
> >>
> >>
> >>>https://glassfish.dev.java.net/issues/show_bug.cgi?id=1059
> >>> Issue #|1059
> >>> Summary|native queries not being committed
> >>> Component|glassfish
> >>> Version|9.0peur1
> >>> Platform|All
> >>> OS/Version|All
> >>> URL|
> >>> Status|UNCONFIRMED
> >>> Status whiteboard|
> >>> Keywords|
> >>> Resolution|
> >>> Issue type|DEFECT
> >>> Priority|P3
> >>> Subcomponent|entity-persistence
> >>> Assigned to|mvatkina
> >>> Reported by|sdo
> >>>
> >>>
> >>>
> >>>
> >>>
> >>>
> >>>------- Additional comments from sdo_at_dev.java.net Thu Aug 31 18:40:04 +0000 2006 -------
> >>>Starting with nightly builds for 9.0_01 this week, native queries are no longer
> >>>being committed. I don't know the exactly nightly it was introduce, but
> >>>somewhere between promoted build 8 and the August 30 nightly build.
> >>>
> >>>See the attachment for the test case. Essentially, we have this code (executed
> >>>within a stateless session bean with default transaction semantics):
> >>>
> >>>Query q = em.createNativeQuery("select object(c) ... for update");
> >>>Object o = q.getSingleResult();
> >>>
> >>>The first time this executes, the row lock is placed on the database, and the
> >>>session bean executes -- this should end the transaction and commit the database
> >>>work. However, that doesn't happen, and hence a second call to the same code
> >>>will hang because it can't access that row in the database.
> >>>
> >>>---------------------------------------------------------------------
> >>>To unsubscribe, e-mail: issues-unsubscribe_at_glassfish.dev.java.net
> >>>For additional commands, e-mail: issues-help_at_glassfish.dev.java.net
> >>>
> >>>
> >>>
> >>>