MaTransaction memory hog

[Igor Stasenko]

2009/10/24 Chris Muller <asqueaker at gmail.com>:
> Earlier this year, I had switched Magma over from requiring
> applications to #finalizeOids to registering with WeakArray.
> Apparently, I missed one!  Thanks, I will fix that.
>
> sig, I have not experienced any problems myself since making the
> above-mentioned change.  There are no race conditions that I am aware
> of, can you elaborate about the problems you experienced?
>

I'm not sure.. it was long ago.
Remember the guy who experienced the problems with getting his objects
out-of-sync with db?

Apparently, the only problem which could occur, that when you in the
middle of finalization, cleaning the dictionaries from
obsolete entries, being interrupted with higher priority process which
by touching the domain objects and triggers modifications to these
dictionaries (like adding new oid's & dirty objects).

> Thanks,
>  Chris
>
> On Fri, Oct 23, 2009 at 4:51 PM, Igor Stasenko <siguctua at gmail.com> wrote:
>> 2009/10/23 Carlo <snoobabk at yahoo.ie>:
>>> Hi
>>>
>>> First off I'd like to say what a pleasure working with Magma has been so far
>>> and I haven't even got to the HA functionality :)
>>>
>>> I noticed in my little app that memory was increasing over a few days. I
>>> think I pinpointed it to MaTransaction not finalizing the readSet.
>>> i.e. in MaTransaction>>initialize
>>>        super initialize.
>>>        readSet := MagmaPreferences weakIdentityKeyDictionaryClass new.
>>>        largeCollectionChanges := WeakSet new.
>>>        self resetCommitPackage.
>>>        ignoreModifiedSignals := false.
>>>        "Conveniently initialized to avoid nil check in code."
>>>        commitNumber := 0
>>>
>>> I modified the initialization of the readSet to be registered with the
>>> finalization process (WeakArray addWeakDependent: readSet.) i.e
>>>        super initialize.
>>>        readSet := MagmaPreferences weakIdentityKeyDictionaryClass new.
>>>        WeakArray addWeakDependent: readSet.
>>> ...
>>>
>>> So far memory has improved and sits around 70mb consistently.
>>> Hopefully this will be helpful for someone else.
>>
>> As far as i remember, Magma using own means to finalize values in
>> readset and rehash it.
>> Registering readset in finalizatin process is a bit dangerous, because
>> we had experienced some problems
>> with it before - i.e. corruption of magma internal structures, because
>> of race conditions.
>>
>>> Cheers
>>> _______________________________________________
>>> Magma mailing list
>>> Magma at lists.squeakfoundation.org
>>> http://lists.squeakfoundation.org/mailman/listinfo/magma
>>>
>>
>>
>>
>> --
>> Best regards,
>> Igor Stasenko AKA sig.
>> _______________________________________________
>> Magma mailing list
>> Magma at lists.squeakfoundation.org
>> http://lists.squeakfoundation.org/mailman/listinfo/magma
>>
>



-- 
Best regards,
Igor Stasenko AKA sig.