RE: Another go at lc75 and lc88 language (correction)

Doug,

Since layering in implementations remains stubbornly non-normative, and if
the algorithm for construction of [message id] is not specified and left to
the imagination, then the temptation will exist to utilize message id in any
way that is useful to the implementers of a higher level protocol.  One
could specify that [message id] shall only be used for correlation, but that
is a bit reaching as well, since ws-addressing does not specify primacy of
authority and over-all architecture is left as a proof for the reader,
besides, users of ws-addressing may already have alternative correlation
capability.  In other words, unless the algorithm is specified, and so far,
except in a proposed example we have not, the issue will remain despite the
anyone's intentions.

thanks

-bob

 

  _____  

From: public-ws-addressing-request@w3.org
[mailto:public-ws-addressing-request@w3.org] On Behalf Of Doug Davis
Sent: Wednesday, June 08, 2005 11:38 AM
To: Robert Freund
Cc: mark.little@arjuna.com; paul.downey@bt.com; public-ws-addressing@w3.org;
public-ws-addressing-request@w3.org; rsalz@datapower.com; tom@coastin.com
Subject: RE: Another go at lc75 and lc88 language (correction)

 


Robert, 
  I get worried when some WS-* spec wants to set some other WS-* spec's
data.  What would do you if WS-AAA has one algorithm, WS-BBB has another and
both specs are used at the same time?  If some other WS-* spec wants some
unique ID for its own purposes its safer if it just defines its own header
and not try to overload or change WS-A's. 
thanks, 
-Doug 





"Robert Freund" <bob.freund@hitachisoftware.com> 
Sent by: public-ws-addressing-request@w3.org 

06/08/2005 06:11 AM 


To

<paul.downey@bt.com>, <tom@coastin.com>, <mark.little@arjuna.com> 


cc

<rsalz@datapower.com>, <public-ws-addressing@w3.org> 


Subject

RE: Another go at lc75 and lc88 language (correction)

 


 

 





If we specify that the message id must be unique, but do not specify the
algorithm, does that not permit other protocols to define a specific
algorithm more to their liking as long as the result remains an IRI?
Would that not supply adequate flexibility?
-bob

-----Original Message-----
From: public-ws-addressing-request@w3.org
[mailto:public-ws-addressing-request@w3.org] On Behalf Of paul.downey@bt.com
Sent: Tuesday, June 07, 2005 9:12 AM
To: tom@coastin.com; mark.little@arjuna.com
Cc: rsalz@datapower.com; public-ws-addressing@w3.org
Subject: RE: Another go at lc75 and lc88 language (correction)


Tom

> I guess what I am seeing in this discussion is that the requirements for 
> wsa:messageId are not clear.

I don't see that at all. I see we have messageId for correlation, 
though I heard during the discussion in Berlin several potential
use-cases enabled by having a mandatory messageId. These included,
but were by no means limited to, transport independent message
correlation, logging and auditing by a transparent proxy observing
messages being exchanged, and elimination of duplicate messages.

> If it is intended to be a general purpose message id for use in wsa 
> message correlation as well as for other uses, then
> we might want to include the uid,integer pair.

I don't see how that follows at all. The introduction of integers 
implies to me you are thinking about some kind of windowing protocol 
where messages may be collated, missing messages may be identified 
and requested to be retransmitted within a range of messages. Possibly
even that the sequence number will roll-over and be reused.

That goes way, way beyond addressing, and isn't something we should
be following at all. From a procedural POV, i'm puzzled as to
what LC issue introducing such complexity and restriction
on the format of messageId would resolve.

> However, if it is just for use in ws addressing correlation, then 
> a URI would suffice, with no semantics on its contents.

I'm OK with a URI, but don't follow the reasoning.

> With a clarification on such a restricted use, it should only be 
> required when the reply to is present with a non anonymous value.

No. I disagree. That's making matters more complex and dependent
upon the binding for a single hop, rendering messagId less useful in 
end to end message passing scenarios where the message passes over
HTTP and then another transport such as Email/MQ.

> Other ws specs which need further semantics would be required to define 
> their own identity elements.  

Possibly, possibly not. It seems to me that making messageId 
mandatory and unique will make it more attractive for other
specs to layer upon it for identifying a unique message. 

> This has the advantage of allowing their 
> use when ws addressing is not being used.

That's their business.

Paul