If you’re using Declarative Services to consume these other dynamic references 
then there is no need to worry. If you’re trying to programmatically write a 
prototype scoped service that has service dependencies then stop and use DS 
instead. Trying to correctly manage the “unget” chains that need to occur when 
one service in a dependency tree is unregistered is just too hard to be 
worthwhile. Tools like the ServiceTracker can make it tractable, but it still 
leaves you writing huge quantities of messy lifecycle code that’s a nightmare 
to debug.

Also, you are correct that you must not keep using a service instance when the 
service has been unregistered. It is your job to discard that reference :).

Tim

> On 23 Aug 2018, at 13:31, Alain Picard <pic...@castortech.com> wrote:
> 
> Just a small note, I should have stated that my worry is about the unget 
> timing. I obviously have a reference to the object and this won't disappear 
> by itself, but if that service has other dynamic references that go away and 
> I keep using the service, I might be in trouble. But I guess the template 
> that I used already had a bit that issue with the supplier (which we seldom 
> use).
> 
> Alain
> 
> On Thu, Aug 23, 2018 at 7:43 AM Alain Picard <pic...@castortech.com 
> <mailto:pic...@castortech.com>> wrote:
> Tim,
> 
> Based on your referenced javadoc, some more googling, I used and adapted from 
> our own current tracker and supplier to create some Prototype versions. Tests 
> are showing correct results, but this is not directly using the 
> PrototypeServiceFactory, so I would appreciate a very quick confirmation that 
> I'm not missing anything.
> 
> Thanks
> 
> Alain
> 
> 
> On Wed, Aug 22, 2018 at 11:54 AM Alain Picard <pic...@castortech.com 
> <mailto:pic...@castortech.com>> wrote:
> Thanks! I actually saw that being called by ComponentServiceObjects while 
> perusing the code.
> 
> Alain
> 
> 
> On Wed, Aug 22, 2018 at 11:52 AM Tim Ward <tim.w...@paremus.com 
> <mailto:tim.w...@paremus.com>> wrote:
> Registering a prototype service is almost as easy as registering a singleton 
> service. Instead of registering a single object you register an instance of 
> PrototypeServiceFactory 
> <https://osgi.org/javadoc/r6/core/org/osgi/framework/PrototypeServiceFactory.html>.
>  This will get called by the framework to get and release instances as needed.
> 
> Tim
> 
>> On 22 Aug 2018, at 16:49, Alain Picard <pic...@castortech.com 
>> <mailto:pic...@castortech.com>> wrote:
>> 
>> Tim,
>> 
>> This helps quite a bit and clarifies a few points for me. As someone who is 
>> migrating from a pre-DS environment and dealing with lots of legacy, how can 
>> prototype scoped services be used outside of DS? That would be fantastic. 
>> Right now we have a good solution to use singleton services outside of DS 
>> but not for "factory" type services.
>> 
>> Thanks
>> Alain
>> 
>> 
>> On Wed, Aug 22, 2018 at 11:27 AM Tim Ward <tim.w...@paremus.com 
>> <mailto:tim.w...@paremus.com>> wrote:
>> Hi Alain,
>> 
>> A "Prototype scoped" service is one where the client(s) can request an 
>> arbitrary number of instances of the “same” service, whereas a 
>> ComponentFactory is a mechanism for the clients to request an arbitrary 
>> number of differently configured component instances.
>> 
>> From the perspective of the component the key difference is that all of the 
>> instances of a prototype scoped component have the same component 
>> properties, and the instances created by the factory component have the 
>> combination of these component properties *plus* the properties passed to 
>> the factory.
>> 
>> In some senses prototype scoped services are better because they:
>> 
>> Don’t require the service implementation to use DS (they may wish to use 
>> something else)
>> Will have satisfied references and configurations (component factories can 
>> be given configuration which invalidates the registration resulting in an 
>> error)
>> 
>> The main reason that you would use a Component Factory rather than a 
>> prototype scoped service is if you genuinely want to have different 
>> specialised configurations for each instance, and it doesn’t make sense to 
>> use a managed service factory (i.e. the customised instances are only 
>> interesting to one client, or must not be shared for some reason).
>> 
>> If your instances are identically configured (or can be, with an init later) 
>> then a ComponentServiceObjects getService() call should be all you need each 
>> time you need a new instance, followed by a call to ungetService() later 
>> when you’re done with it.
>> 
>> Tim
>> 
>>> On 22 Aug 2018, at 12:06, Alain Picard <pic...@castortech.com 
>>> <mailto:pic...@castortech.com>> wrote:
>>> 
>>> On the 2nd part of the question regarding 
>>> ComponentFactory/ComponentInstance vs Prototype/ComponentServiceObjects. I 
>>> get the feeling that CSO should be favored, but I saw an old post from 
>>> Scott Lewis about configuration and that is a bit close to some of my use 
>>> cases.
>>> 
>>> I have cases where I have a Factory component that delivers instances and 
>>> calls an init method to configure the component, or might sometimes return 
>>> an existing matching one that is already cached (like per data connection 
>>> instances). With ComponentFactory I can create a new instance, call init on 
>>> the new instance and return the ComponentInstance. The caller can then call 
>>> getInstance and call dispose when done. I struggle to find a correct/easy 
>>> way to do this with CSO. Am I using the best approach or not?
>>> 
>>> Thanks
>>> Alain
>>> 
>>> 
>>> On Wed, Aug 22, 2018 at 3:46 AM Tim Ward via osgi-dev 
>>> <osgi-dev@mail.osgi.org <mailto:osgi-dev@mail.osgi.org>> wrote:
>>> 
>>> 
>>>> On 21 Aug 2018, at 20:53, Paul F Fraser via osgi-dev 
>>>> <osgi-dev@mail.osgi.org <mailto:osgi-dev@mail.osgi.org>> wrote:
>>>> 
>>>> On 22/08/2018 5:40 AM, Paul F Fraser via osgi-dev wrote:
>>>>> On 21/08/2018 10:00 PM, Tim Ward via osgi-dev wrote:
>>>>>> Have you looked at what the OSC project does? It uses Vaadin, and uses 
>>>>>> the ViewProvider interface to provide view instances. These 
>>>>>> automatically have a detach listener added on creation so that they get 
>>>>>> correctly disposed when their parent container is closed.
>>>>>> 
>>>>>> See 
>>>>>> https://github.com/opensecuritycontroller/osc-core/blob/4441c96fe49e4b11ce6f380a440367912190a246/osc-ui/src/main/java/org/osc/core/broker/view/OSCViewProvider.java#L60-L67
>>>>>>  
>>>>>> <https://github.com/opensecuritycontroller/osc-core/blob/4441c96fe49e4b11ce6f380a440367912190a246/osc-ui/src/main/java/org/osc/core/broker/view/OSCViewProvider.java#L60-L67>
>>>>>>  for details.
>>>>>> 
>>>>>> Tim
>>>>> 
>>>>> Hi Tim,
>>>>> The R7 Spec 112.3.6 states that "SCR must unget any unreleased service 
>>>>> objects" and it sounds to me that the system is supposed to clean itself 
>>>>> up.
>>>>> What am I missing.
>>>> What am I missing?
>>>> 
>>>> Apart from a question mark.. that is.
>>> 
>>> Hi Paul,
>>> 
>>> You are correct in your interpretation of the specification, however…
>>> 
>>> This only happens if you use ComponentServiceObjects, not ServiceObjects 
>>> (which is why this type was added to the DS spec). If you use 
>>> ServiceObjects directly then SCR cannot reference count them and cannot 
>>> help you.
>>> The “leaked” instances are only cleaned up when your component is disposed 
>>> by SCR (for example if it becomes unsatisfied).
>>> 
>>> In this case we *are* using ComponentServiceObjects (good) but we need to 
>>> dispose of the referenced instance when the UI view is closed.
>>> 
>>> If we left it up to SCR to clean up, and our component wasn’t 
>>> deactivated/disposed between UI sessions then we would have a memory leak. 
>>> In general when you use ComponentServiceObjects you should think about the 
>>> lifecycle of the objects you create, and how they are going to be released. 
>>> In this case the component may get an arbitrarily large (and increasing) 
>>> number of instances over time, so it must also dispose of them. If the 
>>> example just grabbed 2 (or 5, or 10) instances at activation and used them 
>>> until deactivation then it would not be necessary to release them (SCR 
>>> would do it for us).
>>> 
>>> I hope that this makes sense,
>>> 
>>> Tim
>>> 
>>> 
>>>>> 
>>>>> Paul Fraser
>>>>> _______________________________________________
>>>>> OSGi Developer Mail List
>>>>> osgi-dev@mail.osgi.org <mailto:osgi-dev@mail.osgi.org>
>>>>> https://mail.osgi.org/mailman/listinfo/osgi-dev 
>>>>> <https://mail.osgi.org/mailman/listinfo/osgi-dev>
>>>>> 
>>>> 
>>>> _______________________________________________
>>>> OSGi Developer Mail List
>>>> osgi-dev@mail.osgi.org <mailto:osgi-dev@mail.osgi.org>
>>>> https://mail.osgi.org/mailman/listinfo/osgi-dev 
>>>> <https://mail.osgi.org/mailman/listinfo/osgi-dev>
>>> _______________________________________________
>>> OSGi Developer Mail List
>>> osgi-dev@mail.osgi.org <mailto:osgi-dev@mail.osgi.org>
>>> https://mail.osgi.org/mailman/listinfo/osgi-dev 
>>> <https://mail.osgi.org/mailman/listinfo/osgi-dev>
> 

_______________________________________________
OSGi Developer Mail List
osgi-dev@mail.osgi.org
https://mail.osgi.org/mailman/listinfo/osgi-dev

Reply via email to