I am not familiar with orchestra, so I can't comment there. It has
not been an option for us up to this point.
However, if you leave a transaction active after a response, it's
always going to be an issue no matter what framework you use.
For us, the problem with holder objects is the deep hierarchy of
entities and relationships we have. We have probably close to a
thousand entities right now, and most multiple-request-spanning
transactions deal with very complex, deep, and width relationship
paths. They only work well in the simple cases where we don't need
them.
On Fri, Jul 2, 2010 at 3:39 AM, Werner Punz <werner.p...@gmail.com> wrote:
> Mike you left out the obvious one, simply use a conversation framework.
> The problem is not transactions but it is that the entity manger is dropped
> along the way hence silently detaching all objects and running you into
> detached error hell. (You still can either setup your jpa provider so that
> lazy loading can happen outside of transaction barriers or prefetch
> everything via fetch join)
>
> I personally found as soon as you go to a conversation framework things
> become way easier (although not entirely easy)
>
> Also what was the problem with holder objects. I personally am thinking of
> moving that way especially since JPA allows to map it transparently via
> queries like following select new FakeHolder(entity.id, entity....) from
> EntityClass entity).
>
> The downside is that you need more logic for pushing the data back into the
> entity objects before writing. But the fake holder pattern is exactly what
> iBatis enforces (although it has the write back logic pushed into the
> configuration) and it works out well in a web centric szenario.
>
>
> WErner
>
>
> Am 01.07.10 23:54, schrieb Mike Kienenberger:
>>
>> I am, sort of.
>>
>> You really can't leave the transaction open beyond the request
>> response as it may never complete.
>>
>> Some of the ways you can deal with it are:
>>
>> 1) work with fake holder entities that get changed back into real
>> entities at the final commit. Very ugly -- tried this one at first,
>> but I don't use it anymore.
>>
>> 2) Work with detached objects. Reattach them back right before the
>> final commit. This is what I currently do. I basically invented a
>> Unit-Of-Work framework that runs over the top of JPA. The unit of
>> work has a separate persistence manager that loads an object, then
>> immediately detaches it. Our framework requires each object to call
>> save() to commit changes. When in the UoW, all save does is add the
>> object to a change-tracker (inserts, deletes, updates). Then when
>> the UoW is committed, the objects are persisted or merged, then
>> committed all in one method call.
>>
>> But in all honesty, this approach also has caused us a lot of hassles.
>> We are most likely going to dump JPA and replace it with Apache
>> Cayenne, which uses a real unit of work concept.
>>
>> Another option for you might be to use an implementation-specific unit
>> of work provided by your JPA implementation. However, I don't know
>> if you might have other issues. I used Cayenne before I used JPA,
>> and I know Cayenne does exactly what I need.
>>
>> A third option you could consider if you want to risk leaving the
>> transaction open.
>>
>> a) Catch the window onunload event, and mixed with ajax, send an ajax
>> request when the user improperly attempts to leave the page (closes
>> the window or browser, enters a url directly, back buttons, some other
>> non-transaction-friendly link clicked).
>>
>> I got this far with that approach, but didn't pursue it. Note that
>> this can only detect when the user is about to leave the page. It
>> cannot do anything at that point -- you'd have to do something about
>> it in some other way.
>>
>> window.onbeforeunload = confirmExit;
>> function confirmExit()
>> {
>> if (needToConfirm)
>> return "You have attempted to leave this page. If
>> you have
>> made any changes to the fields without clicking the Save button, your
>> changes will be lost. Are you sure you want to exit this page?";
>> }
>>
>>
>> b) Some other kind of client-side state tracking so that you know when
>> the user has navigated away from the current multi-request task.
>> We're sticking a taskGroupIdentifier field on every form (ajax
>> included) so we know when the user does something to switch to a new
>> task than the one we're currently working with. Doesn't help if you
>> have some transaction left open and the user never hits the web server
>> again, but a timeout could deal with that.
>>
>>
>> In short, I think it's a difficult problem, and I think JPA is
>> incapable of dealing with it correctly. The other shortfall we have
>> is that JPA cannot rollback a transaction. You have no way of
>> knowing what state your application is in once you roll back the
>> transaction. Again Cayenne automatically puts everything exactly
>> where it was at the start of your unit of work if you do a rollback.
>>
>> On Thu, Jul 1, 2010 at 9:53 AM, Bruno Aranda<brunoara...@gmail.com>
>> wrote:
>>>
>>> Hi,
>>>
>>> Is anyone here using long JPA transactions in their applications
>>> (transactions that span more than one request) but not using Orchestra?.
>>> How
>>> are you doing it?
>>>
>>> Cheers,
>>>
>>> Bruno
>>>
>>
>
>
>
