commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Harish Krishnaswamy <hkrishnasw...@comcast.net>
Subject Re: [Hivemind] ServiceImplementationFactory - no Exception (?)
Date Fri, 19 Mar 2004 02:38:01 GMT
To reiterate what Christian has said, you *will* get a RuntimeException 
if there is no contribution for a service or if the service 
initialization failed. When you will get the exception depends on the 
service model. In the case of deferred services (singleton, pooled and 
threaded models), the initialization of the service is deferred until 
the first access of the service. And that's when you will get the 
exception if there's a problem. Hence it appears as though you have a 
half initialized service while all you have is simply a hollow proxy to 
the service.

Now if you want all your exceptions up front at load time rather than at 
runtime you may mark all your services as primitive (primitive model) 
and you should be fine. But be warned that this may have some unecessary 
resource overheads by way of loading everything at startup whether you 
will be needing them or not.

Now, like I said before, I do agree that we need to throw an exception 
at the end of loading all services if there was a problem and let the 
client decide if they want to continue or kill.

Hope this clarifies.

-Harish

btomasini@neteverything.com wrote:

> Please remember that I have only used HiveMind here for a few days, so 
> feel free to add large grains of salt to these thoughts.
>
> I think I see what you mean.  Two me, there are two sides of this.
>
> 1. On the intial load of modules / services, I would like to see the 
> application have a complete excepion chain to whatever is loading the 
> registry.  In specific, Intializable.initializeService does not throw 
> an exception.  If there is an error in here, it would be nice for the 
> registry to know about it, log it, and invalidate the service.
>
> 2. On the getService side, here are my thoughs.
>
>  (A) If a service fails to initialze in step 1, have getService throw 
> and exception, too.  That is fine with me.
>
>  (B) If a service fails to initialize in setp 1, return null.  This is 
> also fine with me.
>
>  (C) If a service has an error initializing, say in 
> Initializable.initializeServce, why would I want an instance of 
> anything?  To me, a half-initialized service is worse.
>
> I want my service to run 100% out of the box, or have a good strong 
> error, even an NPE from the calling application (B above).  I see no 
> value in pretending everything was ok.
>
>
> On Thu, 18 Mar 2004 12:08:48 -0500
>  Harish Krishnaswamy <hkrishnaswamy@comcast.net> wrote:
>
>> Let me ask you this, how are contemplating handling these exceptions? 
>> Are you going to be catching exceptions when ever you do a lookup or 
>> are you planning on letting these exceptions cascade up to the top 
>> and throw a message to the client. 'Cause either way would be no 
>> different than what it is now, wouldn't you say?
>>
>> -Harish
>>
>> btomasini@neteverything.com wrote:
>>
>>> Great!  I may take my hand at a patch, but I fear am too new to this 
>>> project to get my head around it.  Maybe I will Bugzilla this one if 
>>> I have some time.
>>>
>>> Great product all in all.  BTW, any thoughts for 
>>> Intializable.initializeService() throwing an exception, too?
>>>
>>> I don't want to open pandora's box here, but I hit a similar issue 
>>> with that as well.
>>>
>>> Ben
>>>
>>> On Wed, 17 Mar 2004 23:32:04 -0500
>>> Harish Krishnaswamy <hkrishnaswamy@comcast.net> wrote:
>>>
>>>> Christian is absolutely right, I don't know what I was thinking. 
>>>> The conscious decision was about not throwing the exception at load 
>>>> time and should always expect to see an exception when accessing a 
>>>> service point that has no contribution or had problems loading it. 
>>>> Sorry Benjamin! That should certainly be a bug.
>>>>
>>>> -Harish
>>>>
>>>> Christian Essl wrote:
>>>>
>>>>>> But, if I'm a client and I call registry.getService() I 
>>>>>> absolutely want a
>>>>>> service or an exception! Otherwise I have to stick null checks in

>>>>>> all over
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> I agree. Actullay I was always expecting HiveMind to do that. This 
>>>>> is also more consistent, because if the module or the 
>>>>> ServiceExtensionPoint is not found an ApplicationRuntimeException 
>>>>> is thrown.
>>>>>
>>>>> But now I am a bit confused about where this null comes from. As I 
>>>>> see all ServiceModels - except of primitive - always return a 
>>>>> proxy, independent of what the factory actually returns. And the 
>>>>> primitive ServiceModel (and as I think the other Models as well) 
>>>>> uses AbstractServiceModelImpl.constructCoreServiceImplementation() 
>>>>> which does check for null and throws an ApplicationRuntimeException.
>>>>>
>>>>> Maybe Benjamin you could tell how you got this null. So that it 
>>>>> could be reproduced on the current HiveMind from the cvs. To me it 
>>>>> looks like a bug. Or maybe you used an 
>>>>> ServiceImplementationFactory directly - Spring like?  Or I just 
>>>>> miss something.
>>>>>
>>>>> Thanks,
>>>>> Christian
>>>>>
>>>>>
>>>>> On Wed, 17 Mar 2004 01:34:28 -0500, Geoff Longman 
>>>>> <glongman@intelligentworks.com> wrote:
>>>>>
>>>>>> Wait a sec. Sure log errors when the modules are parsed. That 
>>>>>> makes sense.
>>>>>>
>>>>>> But, if I'm a client and I call registry.getService() I 
>>>>>> absolutely want a
>>>>>> service or an exception! Otherwise I have to stick null checks in

>>>>>> all over
>>>>>> the place.
>>>>>>
>>>>>> Geoff
>>>>>> ----- Original Message -----
>>>>>> From: "Harish Krishnaswamy" <hkrishnaswamy@comcast.net>
>>>>>> To: "Jakarta Commons Developers List" 
>>>>>> <commons-dev@jakarta.apache.org>
>>>>>> Sent: Tuesday, March 16, 2004 10:34 PM
>>>>>> Subject: Re: [Hivemind] ServiceImplementationFactory - no 
>>>>>> Exception (?)
>>>>>>
>>>>>>
>>>>>>> Not throwing an exception is a conscious decision. HiveMind is
a
>>>>>>> microkernal to be thought of as a servlet container - it loads
all
>>>>>>> modules at startup and any problems with the descriptors will
be
>>>>>>> identified and logged at load time but will continue to run.
I 
>>>>>>> can see
>>>>>>> an exception being thrown at load time when there is a problem
but
>>>>>>> certainly disagree with the idea of throwing exceptions at 
>>>>>>> runtime. If a
>>>>>>> service is not loaded properly you have two options - don't care

>>>>>>> or fix
>>>>>>> it and reload it!
>>>>>>>
>>>>>>> -Harish
>>>>>>>
>>>>>>>
>>>>>>> Benjamin Tomasini wrote:
>>>>>>>
>>>>>>>> I have started to use Hivemind and have been successful in
porting 
>>>>>>>
>>>>>>>
>>>>>>> over
>>>>>>>
>>>>>>>> some existing work.  It is quite nice!  Very well thought
out. 
>>>>>>>> Keeping
>>>>>>>> the service / proxy layer in place is cool.
>>>>>>>>
>>>>>>>> One suggestion so far...
>>>>>>>>
>>>>>>>> I had a case where my service object from Registry.getService
came 
>>>>>>>
>>>>>>>
>>>>>>> back
>>>>>>>
>>>>>>>> null.  One of the services used the BuilderFactory.  I was

>>>>>>>> getting a
>>>>>>>> log4j ERROR message, but no exception was thrown to my app.
 It 
>>>>>>>> was a
>>>>>>>> simple runtime error - a typo in a config file.
>>>>>>>>
>>>>>>>> Looking further, I see that
>>>>>>>>
>>>>>>>> in
>>>>>>>>
>>>>>>>> org.apache.hivemind.ServiceImplementationFactory
>>>>>>>>
>>>>>>>> the method
>>>>>>>>
>>>>>>>> createCoreServiceImplementation(....)
>>>>>>>>
>>>>>>>> does not throw an exception or anything.
>>>>>>>>
>>>>>>>> It seems that this prevents calling applications from knowing

>>>>>>>> about
>>>>>>>> problems creating a service.  I could always check for null
in the
>>>>>>>> service object, but this isn't quite right, IMO. Especially

>>>>>>>> with the
>>>>>>>> lazy loading, I think burying any exception here is bad,

>>>>>>>> especially 
>>>>>>>
>>>>>>>
>>>>>>> for
>>>>>>>
>>>>>>>> apps that depend on a large number of services.
>>>>>>>>
>>>>>>>> I would be willing to put some work into this and submit
a 
>>>>>>>> patch if we
>>>>>>>> think we need some exception handling here.
>>>>>>>>
>>>>>>>> Ben
>>>>>>>>
>>>>>>>>
>>>>>>>> ---------------------------------------------------------------------

>>>>>>>>
>>>>>>>> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
>>>>>>>> For additional commands, e-mail: 
>>>>>>>> commons-dev-help@jakarta.apache.org
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>> ---------------------------------------------------------------------

>>>>>>>
>>>>>>> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
>>>>>>> For additional commands, e-mail: 
>>>>>>> commons-dev-help@jakarta.apache.org
>>>>>>>
>>>>>>
>>>>>>
>>>>>> ---------------------------------------------------------------------

>>>>>>
>>>>>> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
>>>>>> For additional commands, e-mail: commons-dev-help@jakarta.apache.org
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
>>>> For additional commands, e-mail: commons-dev-help@jakarta.apache.org
>>>>
>>>
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
>>> For additional commands, e-mail: commons-dev-help@jakarta.apache.org
>>>
>>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
>> For additional commands, e-mail: commons-dev-help@jakarta.apache.org
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: commons-dev-help@jakarta.apache.org
>
>

---------------------------------------------------------------------
To unsubscribe, e-mail: commons-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-dev-help@jakarta.apache.org


Mime
View raw message