Hi,
Hey guys, think a minute. What transaction isolation means is _other_
transactions can't see your work until you commit. Of course _your_
transaction can see everything you've done. If you're working with say the
oracle interactive interface and insert a record in a table, if you query
the table you can see it right away, you don't have to commit first.
_other_ people can't see it until you commit. This has nothing to do with
xa or anything else.
[although there is a related issue with "tightly coupled" and "loosely
coupled" transaction branches in xa: loosely coupled means different
branches of the same transaction can't see each others work. This should
only arise when several transaction managers are using the same global
transaction against the same resource, and the XAResources can't agree that
they are using the same resource. Maybe this is what you guys are thinking
of, but this should only be a potential problem with distributed jboss]
I'm pretty sure someone else complained about this a couple of weeks ago,
although I can't find the reference - they were modifying an entity, then
doing a finder method returning a collection including the "just modified"
bean, except they weren't seeing the modifications.
In my opinion, if this requirement was not present in ejb 1.1, is was a
serious requirements bug, encouraging logically inconsistent behaviour from
the container.
Thanks, Bill.
david jencks
On 2001.06.14 15:49:34 -0400 marc fleury wrote:
> |I have been working with databases for a long time,
> |and particularly with Oracle, I am not aware that this
> |can happen, whether in transaction or not, before
> |commit, nobody can see the updated table period in the
> |database..
>
> well that is my point precisely, it seems to imply that the updates
> "before
> the commit" are seen by connections enrolled in the SAME transaction
> THROUGH
> THE DATABASE.... and frankly I am a bit skeptical as to the level of
> support for these features in the db or even if they exist at all or are
> just "wishful features" (heck they don't even support 2pc and xa
> right)...
>
> so this smells of "teen spirit" to me... by requiring "inflight"
> visibility
> of the changed records they put a difficult requirement on the db drivers
> and I don't see it working well.
>
> They could have just required lock steps, as in first commit the changes
> (and the db can follow that semantic) and then issue your findBy as just
> another query...
>
> I am no db expert (they are rare these days) but it strikes me as a
> misguided requirement.
>
> Bill for example couldn't you get the same functionality with the
> serialiazed commits? Is this functionality that you couldn't get
> otherwise?
> you are the first one to require this feature (and you are savvy enough
> to
> scratch your itch so that is cool but still I wonder...)
>
> but the REAL question is "is it true that you cannot see changed tables
> from
> a connection participating in the SAME transaction that changed it in the
> first place?"
>
> regards
>
> marcf
>
> |
> |I don't know much about XA, it having its own set of
> |rules though.
> |
> |__________________________________________________
> |Do You Yahoo!?
> |Spot the hottest trends in music, movies, and more.
> |http://buzz.yahoo.com/
> |
> |_______________________________________________
> |Jboss-development mailing list
> |[EMAIL PROTECTED]
> |http://lists.sourceforge.net/lists/listinfo/jboss-development
>
>
>
> _______________________________________________
> Jboss-development mailing list
> [EMAIL PROTECTED]
> http://lists.sourceforge.net/lists/listinfo/jboss-development
>
_______________________________________________
Jboss-development mailing list
[EMAIL PROTECTED]
http://lists.sourceforge.net/lists/listinfo/jboss-development
