OASIS Mailing List ArchivesView the OASIS mailing list archive below
or browse/search using MarkMail.


Help: OASIS Mailing Lists Help | MarkMail Help

ebxml-msg message

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]

Subject: Delivery Receipt, NRR and MSG/CPPA/BPSS (mis-)alignment and thelayering mishmash (was jumbled into: reliable messaging - hop by hop)


I read one of your recent notes in which
you present some interesting alternatives,
and I would like to comment on these in the
hopes of aligning BPSS, MSG, and CPPA on
the NRR issues.

David Burdett says, almost as an aside:
"The other alternative would be to provide 
two variations of the Delivery
Receipt, one which indicated that the 
To Party MSH had received and the
second that the To Party Application 
had been notified of the message, but I
don't really want to go there as it 
is definitely additional functionality."

Despite your disclaimer, I think this is
an area we should discuss so that we
can align BPSS CPPA and MSG on NRR.
(Maybe pretty soon I can write a
sentence consisting entirely of

I think that there _could_ be one DeliveryReceipt 
with an extension mechanism
that could be used for all the additional 
conditions that tend to get
tacked on to the DeliveryReceipt: 
One condition would be that the To Party
Application was notified, another would be 
that the syntax had been checked,
a third, that the syntax had been checked 
and the data typing had been
checked and the values were in
allowed ranges, and so on. We would in CPPA
need an enumeration for the main
different extensions, and a general
"namespace supported" element to gesture at
yet unthought of extensions (or whatever
is the extension mechanism du jour).

But strictly speaking,
the functionality in these NRR
extensions is beyond what 
a MSH needs to carry out its main
jobs, even when that job includes
supporting NRR. Let me briefly
expand on this.

NRR is something that
rarely ends up getting uniquely
characterized in terms of what needs
to be _implemented_. We know what we
want to obtain more or less, but can rarely agree
upon what will always be sufficient
to attain it. Because any processing
exception that occurs after the MSH 
finishes its work may provide the To-Party
a legal "way out," NRR always seems to be falling
short of removing all wiggle room.

I seriously doubt that we will remove
all the wiggle room, and that there will always
be some exceptional condition after receipt
and after sending a DeliveryReceipt, that
would result in a processing failure on
the ReceivingParty side (bad syntax,
semantics, and so on). To try to rule out
each of these conditions by checking each of them
before sending a DeliveryReceipt, leads to
endlessly bloating DeliveryReceipt functionality, and
finally a need to be in the position of being able
to produce the Business Response before issuing
a DeliveryReceipt. Clearly this drift leads to a layering

In EDIINT we had very similar requirement
flux. Eventually we just left an open extension
mechanism for receipts, in terms of what
can be added by industry groups. RosettaNet
went through similar struggles in working
out its private/public division.

I personally think it would
be good just to draw a line and say that
"Receipt is obtained when the MSH has persisted the
message payload off the wire," and let the MSH
handle sending the DeliveryReceipt, signed
and including an appropriate hash for the
original message. Nevertheless, adding on an
extension mechanism may appease others (especially
in the BPSS side of things) enough
that we can get all three main groups
(BPSS, CPPA, and MSG) to agree that NRR has been
implemented OK. The problem then will be
in documenting this extension mechanism
so that we can capture publication of and
agreements upon extensions within CPPs and CPAs.
We can then let the CPA adjudicate just how
much "creep" into the application realm occurs
(syntax check on payload, etc).

So, David B's compromise might be a good way
to proceed if the MSH is to provide
NRR functionality. 

An alternative would
be to follow Chris Ferris general preference
to push all NRR all the way back up to the application.
(At least I hope I am not misrepresenting Chris here.)
I think, though, that would make the MSH far less 
interesting as a middleware component module. 
Pragmatically, NRR is not something that anything
but new applications would be able to provide and even
then, it might be difficult to ensure access to the
data elements needed to compute the hash to indicate
being in posession of the original message/payload/whatever.

Moving on, David B. later says:

"I actually think that the second one
[involving an application notification assurance] 
is really an application level acknowledgement."

I think we need to wonder why the application
layer needs an acknowledgement, or why
it needs to get involved in NRR issues.

What is an application level acknowledgement 
as opposed to the application
level response? Why should an application
need to get involved in producing a
receipt instead of its defined application level response?
This seems to be forcing applications to get involved
in detailed delivery related issues, when an application
may only know only about its tasks to be performed
and what do with its results (regardless of how the task got 
queued up to it...). 

I think that the BPSS group needs to explain 
why business applications should at all be in the 
business of worrying about generating acknowledgements 
or receipts. This is a source of layering conflict that is
causing both CPPA and MSG difficulties in understanding
how to finish their own tasks. If there is a
BPSS requirement for NRR, does that mean that
the lower layers are to provide an implementation 
supporting NRR? Or is it an indication that the design
of the business data structures in Requests and
Responses support NRR functionality? Or is something
else meant? CPPA has, I think, mainly been thinking
that a BPSS requirement of NRR was a request that
the underlying layers provide an implementation for
NRR (an implementation that can be a basic or 
a more enriched implementation).

Is an application level acknowledgement 
an application response in addition to the real
application business response or perhaps instead of a 
business response? I think that if a business
layer process thinks it needs such an
acknowledgement for its own purposes, 
then that function should be built into BPSS 
specifications as requirements concerning the 
detailed algorithms, hashes, etc in the structured data 
exchanged in requests and responses. In this
case, NRR will be opaque to the MSH layer,
and there will need to be no agreement on how
NRR is to be achieved/implemented (so DeliveryChannel
characteristics do not need to be assessed
to see whether NRR is supported on a channel.)
This may be consistent with the Chris Ferris option
discussed above; I personally would like to see
a POC demonstrating this approach working interoperably
before allowing it in an approved spec.

Dale Moberg

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]

Powered by eList eXpress LLC