[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [sca-assembly] Pointer to SCA event processing presentation
Jim,
This is a very good discussion.
For some historical context, in OSOA, we started exactly where you have:
using async one-way methods with no exceptions, typing using interfaces,
mapping subscribers to services and publishers to references.
(there were some variations on this, where channels were modeled as
"broker" components, subscribers were modeled as references with
callbacks, autowire was used to avoid explicit wires, etc)
This use of service-reference-wiring model posed several problems. We
had to start special-casing lots of things. The end product was a model
that did not make a lot of sense if you come from the pub-sub world (or
even if you did not). It results in an impedance mismatch. Few
non-exhaustive examples:
1) Subscribers are modeled by services, but the first message (subscribe
message in some systems) has to actually flow from the service to the
reference.
2) Publishers should not know who the subscribers are (or if there are
any) and vice versa. When you explicitly wire services and references,
this disconnected model is broken.
3) A publisher wants to publish by sending a message once. It should not
have to send the same message to N subscriber (eg, in Java I don't want
to see an array of references, on which I have to do a
'ref[i].send(msg)' in a loop).
4) A deployment of additional publishers should not require any change
whatsoever to the existing subscribers and vice versa. But a
service-reference-wiring model does.
5) Interfaces and methods in pub/sub model are artificial constructs.
You would of course process these in your language specific constructs
(such as methods in Java). But events are produced and events are
consumed. There is no action implied whatsoever in the contract. A
consumer may ignore a message. Interface contracts do have semantics. To
use an interface/method launchMissles() you have to read the
documentation for that method to understand what it means.
You can squint a little and work around some of the mismatches. But the
cumulative effect of special-casing everything (kind of, death by
thousand cuts) that results in a model that looks nothing like a pub-sub
model, isn't very useful IMHO.
There are more examples, this is just off the top of my head. I'll
respond to the specific questions/proposal in this thread later today.
Thanks.
-Anish
--
Jim Marino wrote:
> OK. Not having been privy to the initial drafting of the specification
> outside of OASIS (i.e. in OSOA), I may raise some questions that have
> already been answered. But for those who are new to this such as myself,
> I'll ask them anyway.
>
> My first reaction - and this is based on incomplete knowledge - is that
> eventing seems to be dragging in a boat-load of additional concepts and
> technologies. For example, as you mention below, there is the
> possibility that an interoperable eventing protocol may be required.
>
> There are also: consumers, producers, events, event types, EDL,
> channels, filters, and channel promotion. Many of those concepts appear
> to simply mirror concepts we already have. For example:
>
> consumers ----> services
> producers ----> references
> events ----> messages
> EDL ----> a type system such as XML Schema or Java
> channels ----> binding metadata
> filters ----> binding metadata
> channel promotion ----> service/reference promotion
>
> The reason why I asked if JMS could be used to implement eventing is
> that I wanted to see if this apparent duplication of concepts could be
> eliminated by accommodating eventing capabilities using existing
> concepts, in particular bindings. Since JMS appears to capture the
> semantics of eventing, this leads me to believe an "eventing" binding
> and a set of intents could be defined to achieve this.
>
> I noticed the deck contained a slide outlining why these additional
> concepts are needed. Before responding to that, I'll outline what I was
> thinking first.
>
> To start, I was thinking there would be an eventing binding which could
> be configured in the following way:
>
> <component name="MyProducer">
> <implementation.java ..../>
> <reference name="channel1">
> <binding.eventing channel="foo"/>
> </reference>
> <reference name="channel1">
> <binding.eventing channel="bar"/>
> </reference>
> </component>
>
> <component name="MyFooConsumer">
> <implementation.java ..../>
> <service>
> <binding.eventing channel="foo" filter=".."/>
> </service>
> </component>
>
> <component name="MyBarConsumer">
> <implementation.java ..../>
> <service>
> <binding.eventing channel="bar" filter=".."/>
> </service>
> </component>
>
>
> public class MyProducerImpl implements SomeService {
>
> @Reference
> protected MyChannel channel1;
>
> @Reference
> protected MyChannel channel2;
>
> public void doSomething(...) {
> Event event1 == //...
> Event event2 == //...
> channel1.onEvent(event1);
> }
>
> }
>
> public interface MyChannel {
>
> @OneWay
> void onEvent(Event event) ;
> }
>
>
> Based on this example, I've included some comments with respect to the
> slide from the deck mentioned above:
>
> 1. No decoupling
>
> I think the above example is fairly decoupled. It requires a reference
> to be bound (but not wired), which I think is the correct behavior given
> that allowing 0..n would require code to check for null pointers making
> it complex. If there was no consumer, the binding would swallow/ignore
> the event.
>
> 2. Requires interfaces for typing
>
> I'm not sure I follow this. My understanding is that for strongly typed
> languages such as Java, some interface is required. If an interface is
> not required, how is an event sent in Java? Even other language, some
> shared type system or type mapping is required.
>
> 3. Interface/operations have processing semantics, events don’t
>
> I'm not sure I agree with the assumption that interfaces have processing
> semantics. How does MyChannel.onEvent() imply processing semantics? One
> thing that would help is if someone could provide a Java code example
> that did not imply processing semantics so a comparison can be mode.
> Note that the MyChannel example was strongly typed, but the following
> could also be defined:
>
> public interface MyUntypedChannel {
>
> void event(Object event);
> }
>
> 4. Doesn’t allow addition of producers/consumers dynamically
>
> I think the proposal above addresses this concern. It would be up to the
> binding infrastructure to handle this. Note that I think there may be an
> incorrect assumption related to references/wires as well. Wires can be
> updated dynamically.
>
> 5. No ability to set filters
>
> This would be handled in the binding configuration. In the example, I
> used an attribute but it could also be modeled as an element for complex
> expressions. I'm hoping the filter language will be left open as opposed
> to inventing one.
>
> 6. Various issues crop up because of the model mismatch. For example,
> sca:reference(0..n) is mapped to an array in Java
>
> This would be addressed by using a bound required (1..1) reference.
> Also, note that 0..n does not have to map to an array in Java, it can
> also map to certain collection types. Are there other issues along these
> lines?
>
> 7. Subscribers do not offer services they offer callbacks
> 8. Services do not subscribe they make software functionality available
> for others to use.
>
> I think these two are debatable. For example, the JMS binding works
> this way. Why can't we use the same principle and keep things consistent
> with JMS?
>
> 9. A subscribe message will be sent from a service to a reference: a
> direction reversal
>
> I think this is conflating services/references and runtime
> infrastructure. The latter is setting up subscriptions, not a service or
> reference. This is consistent with how the JMS binding is designed.
>
> 10. Changes are hard to manage
>
> Could you provide examples?
>
>
> -------------------------------
>
> What would help me is if someone can provide examples of how eventing
> works in a concrete programming language. My own preference is Java, but
> C++, BPEL, etc,. would be fine.
>
> Thanks for reading this far :-)
>
> Jim
>
> On Aug 27, 2009, at 3:24 PM, Martin Chapman wrote:
>
>> We have not addressed that question yet. The current Assembly spec
>> does state in the conformance section that a web services binding
>> is required, and there is a PR comment on that. Depending on how we
>> resolve that comment, we may have to add an analogous
>> interoperable protocol for eventing (or not as the case may be.)
>>
>> Martin.
>>
>>
>>> -----Original Message-----
>>> From: Jim Marino [mailto:jim.marino@gmail.com]
>>> Sent: 27 August 2009 14:14
>>> To: Martin Chapman
>>> Cc: 'OASIS Assembly'
>>> Subject: Re: [sca-assembly] Pointer to SCA event processing presentation
>>>
>>> OK I'm currently just interested in the API. From your response, it
>>> sounds as if the JMS API can be used to capture the semantics of the
>>> eventing proposal.
>>>
>>> Given that, your response does raise an additional question. Is the
>>> proposal mandating an interoperable wire protocol such as WS-Eventing
>>> or AMQP be used? In other words, would it be considered compliant to
>>> build eventing capabilities using a JMS provider that did not support
>>> an interoperable wire protocol? I'm hoping your answer is "yes" :-)
>>>
>>> Thanks,
>>> Jim
>>>
>>>
>>>
>>> On Aug 27, 2009, at 2:48 PM, Martin Chapman wrote:
>>>
>>>> Everything but a standard wire protocol;-)
>>>>
>>>>> -----Original Message-----
>>>>> From: Jim Marino [mailto:jim.marino@gmail.com]
>>>>> Sent: 27 August 2009 00:06
>>>>> To: OASIS Assembly
>>>>> Subject: Re: [sca-assembly] Pointer to SCA event processing
>>>>> presentation
>>>>>
>>>>> Hi Anish,
>>>>>
>>>>> I have a very naive question about the eventing proposal: is there a
>>>>> capability that cannot be implemented using JMS? My impression is the
>>>>> JMS API has everything that is needed to be the foundation of an
>>>>> eventing system.
>>>>>
>>>>> As an aside, I'm not asking this as a way of implying that JMS should
>>>>> be used and/or required.
>>>>>
>>>>> Jim
>>>>>
>>>>>
>>>>>
>>>>> On Aug 26, 2009, at 6:57 PM, Anish Karmarkar wrote:
>>>>>
>>>>>> Hi Sanjay,
>>>>>>
>>>>>> Here is the link to the presentation:
>>>>>> http://www.oasis-open.org/apps/org/workgroup/sca-assembly/download.php/32653/SCA-
>>>>>>
>>>>> EventingProcessing.pdf
>>>>>>
>>>>>> It does talk about the motivation and has use cases.
>>>>>>
>>>>>> HTH
>>>>>>
>>>>>> -Anish
>>>>>> --
>>>>>>
>>>>>> ---------------------------------------------------------------------
>>>>>> To unsubscribe from this mail list, you must leave the OASIS TC that
>>>>>> generates this mail. Follow this link to all your TCs in OASIS at:
>>>>>> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
>>>>>>
>>>>>
>>>>>
>>>>> ---------------------------------------------------------------------
>>>>> To unsubscribe from this mail list, you must leave the OASIS TC that
>>>>> generates this mail. Follow this link to all your TCs in OASIS at:
>>>>> https://www.oasis-open.org/apps/org/workgroup/portal/
>>>>> my_workgroups.php
>>>>
>>>>
>>>>
>>>> ---------------------------------------------------------------------
>>>> To unsubscribe from this mail list, you must leave the OASIS TC that
>>>> generates this mail. Follow this link to all your TCs in OASIS at:
>>>> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
>>>>
>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe from this mail list, you must leave the OASIS TC that
>> generates this mail. Follow this link to all your TCs in OASIS at:
>> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe from this mail list, you must leave the OASIS TC that
> generates this mail. Follow this link to all your TCs in OASIS at:
> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]