----- Original Message -----
Sent: Monday, July 30, 2001 12:11
PM
Subject: RE: T2: ackRequested
attribute in Via element
Arvola
Setting the ackRequested to Signed or Unsigned is
a decision that the designer (and/or implementer) of the business process
makes when they design or build a business process collaboration or
business process transaction. Factors that need to be considere include
(IMO):
- The natuure of the business process/transaction
- e.g. payments probably need to be secure
- The requirements of the individual trading
partners
This brings up a Messaging Service Interface
question. Does the application (i.e., whatever software is above the MSI
layer) instruct the MSH to send a message reliably, and/or to specify the
use of MSH level acknowledgement, or is it the responsibility of the MSH
to infer from the CPA and such message header elements as Service, Action,
CPAId, etc. the quality of service that should be used for sending a
message, and then set the QualityOfServiceInfo and Via elements
accordingly?
[David
B] I think there are several use cases which are all equally
valid:
1. The MSH looks up the
CPA
2. The Application tells the MSH to send the
data reliably
3. The MSH infers what to do by looking at the
service and action (or other data) in the header and looking up what to do
in some type of rules
database.
Once the CPA knows what to do it should set the
QualityOfService and Via elements
accordingly.
I see. The Message Service Interface will
have to be designed to support the different use cases. Perhaps you mean
"Once the MSH knows" in your above sentence.
I have always thought that the ackRequested
attribute should be set if reliable messaging is being used. Is it
allowable to set deliverySemantics to OnceAndOnlyOnce and to either omit
the Via element or to include a Via element but set ackRequested to
None?
[David
B] As currently specified I think that you have to use the Via
element if you are doing reliable messaging as you need ackRequest to
be set even if you are doing a single hop. The reason that ackRequested is
in the Via element is because the need for ackRequested is removed if
you are doing multiple hops and one of the hops is using a proprietary
"reliable messaging protocol" e.g IBM MQ Series. This really indicates
thatr the Via element is wrongly named IMO. What are your thoughts about
changing it ...
The way I read
from the current spec, the sender only needs to set ackRequested
if it is expecting to get back an Acknowledgement message that
actually contains the Acknowledgement element (which carries non
repudiation information). Otherwise, it seems rather redundant to require
both deliverySemantics set to OnceAndOnlyOnce and ackRequested set to
something other than None in order to elicit Reliable Messaging
behavior.
In general, I am
of the opinion that the Via element should be used only when an
intermediary is involved. Therefore, I don't think that the ackRequested
and syncReply attributes belong only in the Via element. Perhaps the
QualityOfServiceInfo element is a more appropriate home for these. Why
should ackRequested be different from hop to hop
anyway?
The concept of an Acknowledgement message to
support reliable messaging seems muddled with the concept of the
Acknowledgement element which is primarily used to provide non-repudiation
of receipt. My colleagues at TIBCO have pointed out to me that it seems
possible to send an Acknowledgement message that does not carry an
Acknowledgement element, and that only the RefToMessageId is
required. (See lines 1816 and 1817 in Section 10.3.3.) Can you
please confirm that this indeed is the case?
[David B] I think that what you say is
correct.
I think what would be really useful is to have a
guide that describes how to design a business process/transaction using
ebXML Messaging. Do you agree? If so hould it be in the 1.1 spec or
something separate.
Either way is fine with me, as long as the guide
is available around the time the 1.1 spec is published.
[David B] Any
volunteers?
I will be happy to contribute to an effort
headed up by some experienced veteran of the MSG spec.
I think that if an error is dicovered then
including the ackRequested set to true on the error message runs the risk
of a never ending series of messages. The only use cases to consider are
where a message is being sent reliably in which case
...
My question was whether the error message should
be sent reliably. I was under the impression that the CPA determines the
role played by the receiver and the delivery channel that it will use for
receiving messages. If the delivery channel uses a doc exchange that calls
for the use of reliable messaging, then shouldn't the error message be
sent reliably?
[David B] Yes if the messages are being
sent synchronously.
- If the message that was in error has
ackRequested set to Signed/Unsigned and the error message sent in return
is lost, then the sender of the original message will resend it which
will cause the error message to be resent - see example 1
below
- If the message that was in error has
ackRequested set to None (e.g. it is a synchronous resposne) then
sending the error message with Ack Requested set to Signed/Unsigned
makes sense otherwise the sender of the error message will not know if
the message was delivered - see example 2 below
EXAMPLE 1
Message (with
error)(AckRequested=S/U)---------------------->
<---------------ErrorMessage (Includes
Acknowledgment element)
If the message is in error, I doubt if the
responder is obligated to include an Acknowledgement element in the error
message, even if AckRequested has been asked for.
[David B] Perhaps. Even
if he did not then the rules say that any errors in an error message
are not responded to so it would not make much
difference.
EXAMPLE 2
Message (no
errors)(AckRequested=S/U)------------------------>
<-------Message (with error) (Includes
Acknowledgment element)
Error Message
(AckRequested=S/U)----------------------------->
<---------------Message (Includes
Acknowledgment element only)
I still
don't see why in example 1 the sender of the error message does not set
AckRequested to S/U while in example 2 the sender of the error message
sets AckRequested to S/U.
[David B] In example 1, suppose the ErrorMessage was
lost, then, as the original message was sent reliably, the sender would
resend it which would cause the Error Message to be resent. This means
that the ErrorMessage does not need to be sent reliably as the sender of
the Error Message knows that they will receive the original message again
if the error message is lost
In example 2, the Error
Message needs to be sent reliably as the sender wants to make sure that it
was received. If it is not sent reliably then there will be no
acknowledgement.
Perhaps the design guide you
suggested earlier should clearly define the error scenarios when reliable
messaging and/or AckRequested are to be used.
A general
rule (it's somewhere in the spec but I can't immediately find where) says
that if you find an error in an error message then you don't respond with
another error message and sort out the problem by some other
means.
Regards
David
Section 8.7 does not clearly indicate the
circumstances under which the ackRequested attribute should be set (to
Signed or Unsigned). Is this governed by the ReliableMessaging and
NonRepudiation element for the DocExchange associated with the
DeliveryChannel that is being used?
In particular, when an error is encountered
in processing a message, what should be the strategy for setting the
ackRequested attribute in the error message? In other words, under what
circumstances, if any, are error messages to be sent
reliably?
Thanks,
-Arvola