It's not that expensive (each context attaches to a shared stack) and
looks like the best strategy in your case.
One possible optimization is to use a single shared DataContext for
read-only operations, and create a new DataContext for each operation
that is expected to modify and commit the objects. It is only possible
of course if you have any "read-only" operations at all.
Andrus
On Jan 7, 2008, at 5:30 PM, Álvaro Martínez wrote:
>
> Thanks, Andrus and Philip
>
> The threads I'm talking about are created from many sources and for
> different reasons. Not of all them are triggered in response to
> "something". There are also watchers, periodic tasks... So I can't
> map data contexts to some particular condition.
>
> So then I have to create one data context per operation (that means
> a set of actions). Is this expensive? We are developing a heavy
> loaded cluster of servers, so it's important.
>
> Thanks again!
>
>
> Andrus Adamchik escribió:
>> Hi Álvaro,
>>
>> It is hard to give a precise advice on multithreading without
>> knowing the nature of your application. So here is a few general
>> notes:
>>
>> * DataContext instance is your isolated area for making in-memory
>> changes to objects that will all be committed at once. So consider
>> using multiple contexts as appropriate. Cayenne docs recommend
>> various common patterns, such as DataContext per session (i.e. each
>> user has a dedicated context), DataContext per request, or
>> DataContext per application (in a read-only app). You can also
>> devise your own approach, if none of the above fit your needs. All
>> you need to know here is that multiple threads *reading* from a
>> shared DataContext is ok, but multiple threads *writing* to a
>> shared DataContext is not ok.
>>
>> * In a rare case if you really need multiple threads to work off of
>> the same context, consider using a dedicated nested DataContext for
>> each atomic object modifications.
>>
>> Andrus
>>
>>
>> On Jan 7, 2008, at 2:44 PM, Álvaro Martínez wrote:
>>>
>>> Hi, I've been working for a while with Cayenne but never realized
>>> I had a problem... until I got a weird exception.
>>>
>>> The fact is that I had been using context.newObject() and
>>> context.commitChanges() to create new rows in the database. But my
>>> application works with many threads, so global commits can (and in
>>> fact do) interrupt normal creation of objects. Thread A and Thread
>>> B are creating objects and filling their fields, but then B
>>> commits all and A throws a validation exception because mandatory
>>> fields are missing.
>>>
>>> How could I commit only one object?
>>>
>>> Thanks,
>>>
>>> Álvaro from Spain (Push the button Inc.)
>>>
>>>
>>>
>>
>>
>>
>
>
This archive was generated by hypermail 2.0.0 : Mon Jan 07 2008 - 10:41:36 EST