David,
But if the CPA is not used, the two parties have to first get on the phone
to negotiate what they have to agree on to be able to send that message
without a CPA, and then they both have put that information into their
systems and get it consistent and correct. There's no free lunch. If the
negotiable information is simple, the CPA will also be simple. In some
cases a third party may be able to dictate all of the variables. The RNIF
seems to be one such case and I assume that CommerceOne is another.
Relying on such third parties may only serve to divide the world into
islands that can't easily interoperate.
I have been discussing this with someone who is trying to implement a
message service to work either way. That's not easy. In order for the
message service spec to cover both cases, some of its statements are so
broad or vague that they don't give sufficient guidance to the implementer.
The CPPA team has an open work item to write an appendix for its spec
showing how the elements in the message header are to be used with the CPA.
However, that information really should be in the message service spec.
Yes, the spec needs to be clarified. Clarification is needed not just to
make it clear that the CPA is optional. All of the ambiguities resulting
from the CPA needing to be optional have to be located and clarified in the
form "this is what you do if you have a CPA and this is what you do if you
don't have a CPA."
Regards,
Marty
*************************************************************************************
Martin W. Sachs
IBM T. J. Watson Research Center
P. O. B. 704
Yorktown Hts, NY 10598
914-784-7287; IBM tie line 863-7287
Notes address: Martin W Sachs/Watson/IBM
Internet address: mwsachs @ us.ibm.com
*************************************************************************************
"Burdett, David" <> on 07/17/2001 01:24:00 AM
To: Arvola Chan <>,
cc:
Subject: RE: Some Issues with the Reliable Messaging Specification in Vers
ion 1.0 of ebMS
Arvola
You have spotted one of the inconsistencies that, in my opinion, needs to
be fixed. It arose because there were two different views on the
relationship between the CPA spec and the ebMS spec. This really boiled
down to whether the CPA was always required or optional. I will say now
that I was always firmly in the latter camp and think that the CPA should
be optional. The main reason for this was that if the CPA is required,
then you cannot just send a message to someone, you have to negotiate the
CPA first and how you do this has not been specified.
You are also right that there are issues around using the reliable
messaging spec over multiple hops which need to be resolved.
However the spec does need to be clarified in my opinion to make it clear
the the CPA is optional. Removal of this inconistency (and a few others)
is one of the things that I will suggest at the ebMS meeting tomorrow and
Wednesday.
Thanks for your comments.
David
PS Most specifications seem to go through some type of revision after they
are published ... ;)
-----Original Message-----
From: Arvola Chan [mailto:]
Sent: Sunday, July 15, 2001 10:43 AM
To:
Subject: Some Issues with the Reliable Messaging Specification in Version
1.0 of ebMS
Section 10.2 in the ebMS spec describes the reliable messaging parameters:
deliverySemantics, mshTimeAccuracy, TimeToLive, reliableMessagingMethod,
ackRequested, retries, retryInterval, persistDuration. Only
deliverySemantics, TimeToLive, reliableMessagingMethod, ackRequested can
be found in Appendix A (ebXML SOAP Extension Elements Schema); retries,
retryInterval, and persistDuration can only be found in Appendix D of the
ebCPP spec.
I find the statement on line 1695 "This parameter information can be
specified in the CPA or in the MessageHeader (section 8.4.2)." imprecise
in the following sense:
mshTimeAccuracy is neither in the MessageHeader nor in the CPA.
TimeToLive is not mentioned anywhere in the CPA. It is not clear how
the sending MSH should pick a value for this parameter.
The sub-sections describing retries, retryInterval, and persistDuration
do not clearly indicate that these parameters are to be obtained from
the CPA. Furthermore, their spellings do not match those in the CPA
(case difference). It will be helpful to specify how elements in the
MessageHeader can be used to look up these values from the CPA. The
CPAId alone is not sufficient. The Service and Action elements will
also have to be used to locate the relevant DocExchange, ebXMLBinding,
and ReliableMessaging elements from the CPA.
It is not clear if the current Reliable Messaging specification works over
multiple hops. Line 1774 prescribes that a TraceHeader be created in
accordance with Section 8.5.2. The latter section however does not say
anything about how to determine the next intermediary, in those cases
where one or more intermediaries are to be involved. The descriptions on
lines 1825 and 1829 on how to populate the From and To element in the
Acknowledgement Message also do not clearly explain the circumstances
under which sub-elements under the last TraceHeader in the incoming message
are to be used.
I also find the following issues with Section 10.3.2 on Receiving Message
Behavior:
Line 1783 uses the URI
http://www.ebxml.org/namespaces/messageService/MessageAcknowledgement.
This is not consistent with the URI specified on line 1823:
uri:www.ebxml.org/messageService.
Steps 2.d)i) on line 1800 and 2.d)ii) on line 1802 are confusing. The
phrase "and resend it" on line 1800 should be struck out. Otherwise,
the message would be resent twice.
Step 2.c) omits the crucial action of passing on the message which has
been found not to be a duplicate to the application.
Step 2.d)iii) is incomplete. The action to be taken when syncReply is
set to True and the CPA indicates no application response is included
is left unspecified. I believe in this case an Acknowledgement Message
should be generated and returned synchronously.
Under Section 10.3.5 Duplicate Message Handling, I find the description of
an "identical message" puzzling. Why is it possible for a duplicate
"identical" message to have an additional TraceHeader? Is the sending MSH
required to append another TraceHeader when it resends a message because
an Acknowledgement Message has not been received in time? Or is it
required to update the Timestamp in the TraceHeader to reflect the time of
retransmission?
Thanks,
-Arvola
Arvola Chan ()
TIBCO Software (on loan to RosettaNet)
+1-650-846-5046 (US-Pacific)
