Re: [sca-assembly] TIBCO eventing proposal

[Anish Karmarkar <Anish.Karmarkar@oracle.com>]

More comments inlined as <anish>...</anish>

Thanks.

-Anish
--

On 11/30/2009 11:13 AM, Scott Vorthmann wrote:
> Responses inline, Anish...
> 
> Scott
> 
> 
> On Nov 25, 2009, at 11:45 PM, Anish Karmarkar wrote:
> 
>> Thanks for sending this out. Even though pictures/examples are missing, I *think* I get the general idea. Few questions/comments about the details (not the overall design) that immediately came to mind:
>>
>> 1) Section 3.1 Exposing a Service as a Consumer says "The name of such an operation is understood to define an event type ..."
>> Can you elaborate on why the operation name has to define the event type? Perhaps I misunderstood what the doc says, but from an event processing perspective, I don't think you want to restrict which events are delivered to a method based on its method name. For example, an onMessage(...) method may (and typically does) processes any event that is delivered. The name onMessage has no relevance wrt which event types it wants to consume.
> 
> This is a question of programming style.  The self-dispatch style of event handling you mention should not be ruled out, but as a developer I would prefer to write to a pre-dispatched interface, "printerOn(...), printerOff(...), etc.".  JAX-WS considers this style appropriate and developer-friendly, after all.
> 

<anish>
I don't have problem with using JAX-WS style of event dispatch. But the 
assumption that the operation name *necessarily* stands for event type 
is problematic. Eventing processing sometimes requires that the 
application be subscribed to all events in a channel and do its own 
dispatch. Furthermore, if we want to enable legacy systems, which I 
would argue we do, we have to deal with the fact that event instances 
don't necessarily have a event type metadata associated with it. Perhaps 
you are thinking of synthesizing the event type metadata for such instances.
</anish>

> If the Java TC wishes to define annotations to support one-method, self-dispatching consumers, that's fine, but we don't believe it is necessary, given the existing mechanisms for one-way operations with JAX-WS.
> 
>> 2) Section 3.2 Exposing a Reference as a Producer does not say anything about the cardinality of the reference. I assume it is 0..n.
> 
> Actually, I expect it is always 0..1 for one-ways.  There are no responses coming back, so no use-case that makes sense to me for 0..n.
> 

<anish>
Having 0..1 does solve the problem of getting an array in the Java code 
and having to send the same message (for a producer) to every member of 
the array. However, assuming the same wiring rules as that for 
service/reference apply, 0..1 prevents a consumer/producer from being 
connected to multiple channels. This disallows important usecases. For 
example, a producer may be producing earthquake related events and the 
producer may have to be connected to the 'data gathering/archival' 
channel as well as the 'tsunami warning channel'.
</anish>

>> 3) WRT the minimum-novelty goal, I should point out that the W3C WSRA WG that is standardizing WS-Eventing and friends has found the need to separate the event type description from WSDL. See http://www.w3.org/2002/ws/ra/edcopies/wseventing.html#Advertising  There is also a plan (not sure if it is approved by the WG yet) for refactoring the event description part in a separate spec in the WG. Trying to fit/overlay event description into the WSDL portType/method model does have its problems.
> 
> Perhaps.  We should discuss those problems.  Our main point is this: even if we deem it useful to define a specific metadata model for event types, we should fit it into our existing extension point for interface metadata, and define a mapping to/from WSDL.
> 

<anish>
Defining a mapping to/from WSDL seems fine and hopefully WSRA will 
tackle that. But I don't see why we have to "... fit it into our 
existing extension point for interface metadata ..."
</anish>

>> 4) In Section 7 Domain Channels, is your objection to the syntax or to the fact that a pub/subcriber deep down in a nested composite is "connected" to a domain-level channel, i.e. visibility issue?
> 
> The latter, visibility.
> 
> 
>> 5) WRT channelBinding.sca, is that really necessary given that binding.sca is magic anyway? Wouldn't binding.sca fit the bill?
> 
> The issue here isn't the semantics, rather just the XML mechanics (or UML if you like).  From a modeling and schema perspective, "bindings" live on references and services.  We need something else to live on channel, consumer, producer.
> 
>> Thanks!
>>
>> -Anish
>> --
>>
>> On 11/25/2009 5:13 PM, Scott Vorthmann wrote:
>>> Here is our proposal, in not a great deal of detail.  We'll have pictures and examples next week.  ;-)
>>> Scott
>>> ------------------------------------------------------------------------
>>> ---------------------------------------------------------------------
>>> 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 
>