RE: [sca-bindings] NEW ISSUE 7: JMS bindingType and ordered intent

["Michael Rowley" <mrowley@bea.com>]

I agree that this needs to be clarified.  There are many possible
semantics that could be associated with the ordered intent.  It also
could imply routing decisions -- when there are multiple consumers
pulling off the same queue, do you want 2 messages that are sent with a
strict ordering to be handled by the same consumer?

I believe that the ordering semantic should be quite weak, so that it
can apply to many possible mechanisms.

Here is a rough proposal:

Under the following conditions:
- Two messages are sent through a single reference.
- The sending component knows the order in which the messages were sent
(i.e. they weren't sent from parallel threads -- this condition needs to
be formalized).
- The two messages are both received by a component's service, and have
not been received in parallel threads (again this needs to be
formalized, or we may want to consider saying that the messages must not
be received in parallel).

Guarantee:
- The two messages will arrive at the receiving component in the same
order in which they were sent.

Naturally, if the definition is in terms of any two messages, the
complete sequence will be ordered by induction.

Michael


-----Original Message-----
From: Eric Johnson [mailto:eric@tibco.com] 
Sent: Friday, October 05, 2007 12:44 PM
To: Peshev, Peter
Cc: sca-bindings@lists.oasis-open.org
Subject: [sca-bindings] NEW ISSUE 7: JMS bindingType and ordered intent

Logged as: http://www.osoa.org/jira/browse/BINDINGS-7

-Eric.

Peshev, Peter wrote:
> TARGET:
> JMS Binding Specification Version 1.1, Working Draft  25 September
2007
>
>
> DESCRIPTION:
>
> The current bindingType of the jms is defined as :
>
> <bindingType type="binding.jms" alwaysProvides="jms"
> mayProvide="atLeastOnce atMostOnce ordered conversation"/>
>
> A guidance for the implementation could be made for the ordered intent
> and what does it mean in the JMS case. 
>
> If it is provided by the assembler, it is especially interesting what
> happens if there are several instances of service consumers sending
JMS
> messages in concurrent. Does the intent means that the messages are
> "ordered" within the instance that produced them - i.e. each service
> consumer instance is having a separate JMS session and in that case
the
> ordering is guaranteed by the semantics of the JMS session.
>
> Or there is a global ordering within all the concurrent service
> consumers ? I.e. there is a global synchronization in the JVM for the
> whole wire, and there is a single JMS session behind it.
>
> Of course if there is no such "ordered" intent, the binding is free to
> do whatever it wants - a pool of JMS sessions, each time opening new
> one, etc.
>
> PROPOSED SOLUTION
>  Include clarification in the spec
>
>
>
>
>