Subject: RE: Yet more issues to resolve
Marty
I honestly don't think that it violates anything in the spec since resending
over a different transport has long been thought of as a good idea. I agree
though that an implementation decision could be to design/implement a
solution that has different MSH for each transport protocol.
In my view, it is more of a CPA/CPP issue, in that if you define alternate
transport channels for a particular message in a business process, then you
need to make sure that you have the technology to support the alternative
transport within the same message. If this is not the case then don't put it
in the CPP/CPA.
Another way you might be able to do it is by sending the same payload again
using a different messageId.
Thoughts?
David
-----Original Message-----
From: Martin W Sachs [mailto:mwsachs@us.ibm.com]
Sent: Friday, January 05, 2001 7:25 AM
To: Burdett, David
Cc: Doug Bunting; 'Prasad Yendluri'; 'Stefano POGLIANI';
ebxml-transport@lists.ebxml.org; ebxml-tp@lists.ebxml.org
Subject: RE: Yet more issues to resolve
David,
>>>It can't always be "exactly the same stream of bytes on the wire since,
if you resend a message using a different transport protocol, then you
need to use a different Routing Header with different URIs.<<<
MWS: We need to be very careful about resending a message using a
different transport protocol. A different transport
protocol may mean a different physical port on the sending machine and
possibly a different instance of the MSH. So I agree that in this case, it
won't be exactly the same stream of bytes. But if it is a different
instance of the MSH, does this violate any assumptions that are behind the
specification? Do we have to beef up the specification to ensure that
multiple instances of the MSH can be involved in the protocol for the same
message?
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" <david.burdett@commerceone.com> on 01/05/2001 02:47:07 AM
To: Doug Bunting <Doug@ariba.com>, "'Prasad Yendluri'"
<pyendluri@webmethods.com>
cc: "'Stefano POGLIANI'" <stefano.pogliani@sun.com>, Martin W
Sachs/Watson/IBM@IBMUS, ebxml-transport@lists.ebxml.org,
ebxml-tp@lists.ebxml.org
Subject: RE: Yet more issues to resolve
Doug
See comments below.
David
-----Original Message-----
From: Doug Bunting [mailto:Doug@ariba.com]
Sent: Thursday, January 04, 2001 11:29 AM
To: Burdett, David; 'Prasad Yendluri'
Cc: 'Stefano POGLIANI'; Martin W Sachs; ebxml-transport@lists.ebxml.org;
ebxml-tp@lists.ebxml.org
Subject: RE: Yet more issues to resolve
Two additional comments related to persistent storage with respect only to
implementing Reliable Messaging. Again, legal or business reasons for
additional storage must be considered and described separately, I think
that's the purpose of the persistDuration parameter.
Persistent storage at the receiving side MUST (in RFC 2119
terms) include both the messageID value and the corresponding
intermediate ACK (if any). The exact same answer should go back to the
sender no matter how many times they send a duplicate message.
>>>I agree, that MessageId needs to be persisted. However persisting
the intermediate ack is a function of the sending process not the
receiving process. See also section 10.2.1.2<<<
If it isn't already clear in the specification (and, I don't think it
is), we should make sure the messageID corresponds to a particular set
of bytes. I believe we've said that retries must use exactly the same
message as before. I don't think we've made "exactly" quite clear
enough. Is it "semantically identical content" or "matching canonical
form" or (my choice) "exactly the same stream of bytes on the wire"?
>>>It can't always be "exactly the same stream of bytes on the wire since,
if you resend a message using a different transport protocol, then you
need to use a different Routing Header with different URIs.<<<
We may also want to describe a possible error (detected by the receiver)
when the checksum of the message doesn't match that (optionally) stored
along with a messageID.
>>>One of the problems of doing checksums, is that it has to be done over
a canonical version of the document. Sending the same bytes might result
in a different set of bytes being received, especially if you are using a
transport such as SMTP that can alter the message content. So I think that
if this is important, then you should use XML Dsig to make sure that the
message is the same as it includes transformations that allow a canonical
form to be generated. <<
I definitely agree that the sender needs to persist the entire request
until it receives the intermediate or final acknowledgement. The receiver
should not be required to persist the request for reliable messaging
purposes, though they could. (Of course, a receiver that is forwarding
the message in turn must persist the request as any other sender would.
Since we distinguish between the inbound and outbound MSH instances at
intermediate nodes, this might not need to be described in our
specification.) The intermediate choice of storing a hash or
checksum would be OPTIONAL (also in RFC 2119 terms).
>>>Please see comments above, and the revised definition of
PersistDuration<<<
thanx,
doug
-----Original Message-----
From: Burdett, David [mailto:david.burdett@commerceone.com]
Sent: January 3, 2001 21:56
To: 'Prasad Yendluri'
Cc: 'Stefano POGLIANI'; Martin W Sachs; ebxml-transport@lists.ebxml.org;
ebxml-tp@lists.ebxml.org
Subject: RE: Yet more issues to resolve
Prasad
See comments below.
Regards
David
-----Original Message-----
From: Prasad Yendluri [mailto:pyendluri@webmethods.com]
Sent: Wednesday, January 03, 2001 7:21 PM
To: Burdett, David
Cc: 'Stefano POGLIANI'; Martin W Sachs; ebxml-transport@lists.ebxml.org;
ebxml-tp@lists.ebxml.org
Subject: Re: Yet more issues to resolve
David,
The description for the "persistDuration" parameter reads as follows:
PersistDuration is the minimum length of time in days that a Message that
is sent reliably is kept in Persistent Storage by a MSH. The value used
for PersistDuration is an implementation decision although it MUST be
greater than the value of the TimeToLive parameter for any message that is
sent.If a duplicate message (i.e. with the same MessageId ) is received
before the PersistDuration has passed, then the MSH that receives it MUST
process it as a duplicate message as described in sections 10.2.1.3 and
Error! Reference source not found..
If a duplicate message is received after the PersistDuration has passed,
then although it may be treated as a duplicate, the sender must realize
that it will probably be treated by the MSH as if the message were a new
message that had not been received before.
1. If the purpose of this is to simply detect duplicates why require the
entire message to be stored? Isn't saving the value of the messageId
enough?
<DB>
Yes, the MessageId is enough. It was not my intention to REQUIRE that a
whole message should be kept. Note that I did not say that "
PersistDuration is the minimum length of time in days that a **complete**
Message that is sent reliably **MUST be** kept in Persistent Storage by a
MSH. Also lines 1213-1216 in Reliable messaging, that describe the
behavior of a MSH that receives a message, explicitly say ...
>> 1) If the message is not a duplicate then do the following:
a) Save the MessageId of the received message in persistent
storage. As an implementation decision, the whole message MAY be stored if
there are other reasons for doing so.<<
Another factor is that, if you want to support a Message Status Inquiry,
then you will also need to record the time a message was received. I agree
though that the description of persist duration could be clarified to
indicate that only the MessageId needs to be saved.
</DB>
If any, I would think it is the sender MSH that needs to persist the
message, so that a retry would a simple retransmission.
<DB>
I agree the sender MSH does need to persist the message. Line 1200 in the
Reliable Messaging specification that is describing the behavior of a
sender of the message says ...
>> 1) Save the message in persistent storage (see section 10.1.1)<<
However how long the sender saves the message for is not relevant to
Reliable Messaging. You need to know it for the receiver since you need to
know if you resend a message whether or not duplicate filtering will work
which it won't if the reciever has not persisted the message. The same
does not apply to the sender since the persisting of the message only
applies to resending.
</DB>
2. Why the value limited to "days"?
<DB>A good point. My original thinking was that if the message is being
placed in some type of persistent storage such as a disk, then you
probably would not want to clear it out too often. So setting it to whole
days makes it a simpler value than the years, months, weeks, days, hours,
minutes, millisecs?? that you might otherwise need. Another approach is to
specify a single integer and then put the units in separately such as
<persistDuration units="seconds">20</persistDuration> in the CPA. Would
that work better?</DB>
3. The last paragraph seems to imply that sender and receiver could be in
an inconsistent state without clear knowledge of, if retries on a message
are still permitted or not. That is sender thinks retries are still
permitted but, receiver does not know it is a re-send, as the message was
deleted from the persistent store. If we got the Reliable Messaging
specification right, this should never be the case. At least one of the
parties should always know. The receiver must always know if the received
message is duplicate or not, if the message was received once
successfully, irrespective of, if the entire message is (still) persisted
or not (based on the MessageId alone). We should never retire the
MessageId either from sender/receiver (or intermediary) perspective. It is
a GUID isn't it?
<DB>I agree that ideally the Message Ids should never be "retired".
However what I think you are suggesting is that the MessageIds are
**never** deleted by the receiver. I honestly don't think this is
practical, since if you want to forever to be able to do a duplicate
check: a) the MessageIds received have to be immediately available in
order to provide a rapid response to the duplicate check, and b) this
would imply ever increasing persistent storage which would prove costly to
maintain. This means that the recipient of a message will have to, at some
point, delete old message ids. The point is that the sender needs to know
whether or not duplicate filtering will work. By specifying a
persistDuration for received Message Ids, the Sender will knowthat the
duplicate filtering will work, especially if the persist duration is
significantly longer than any period of time over which a message might be
resent. However the sender can never know EXACTLY when the sender will
delete a message id unless they do a query first.
Perhaps it would be better if the last paragraph said ... "A MSH SHOULD
NOT resend a message to a receiver if the receiver has probably deleted
the message since the time indicated by PersistDuration has passed since
the sender first sent the message." You could even require that a sender
report a delivery failure if the time has passed.
</DB>
4. I think, how long a message should be persisted in the MSH should be
based on acknowledgment and retries and associated timeout constraints of
the business process choreography and not based on a separate parameter
that tells how long the message should be kept. Any parameter that
dictates how long a message is persisted should probably be driven by
legal requirements such as non-repdudiation.
<DB>There has been recent discussion on this topic on the list. See the
thread at ...
http://lists.ebxml.org/archives/ebxml-transport/200012/msg00211.html
I disagree that basing persistence duration solely on acknowledgement and
retries is sufficient. This can work for a sender if there are no legal or
other reasons for saving the message. However it is insufficient for a
receiver of a message.
I completely agree though that the value set on message persistence by a
recipient of a message is governed by several factors including:
a) the need for a sender to know if reliable messaging will work
b) the requirements of a business process choreography
c) other needs such as legal requirements.
This is why persistDuration is a parameter that can only be set in the
CPP/CPA (see the table at line 1482). It can't be set in the header.
Therefore persistDuration can be, as you suggest, controlled by the needs
of the business process as the CPA much of the information in the CPA is
controlled by business process needs. However, the information still needs
to be known by the MSH that is doing reliable messaging and hence the need
to include as a CPA parameter that is used by the MSH.
</DB>
Regards, Prasad
"Burdett, David" wrote: Stefano
Version 0.91 of the spec has the concept of a "persistDuration" parameter
that would be in the CPA/CPP that identifies the number of days a message
that is sent persistently should be kept by the receiving MSH.
The spec is not specific about how long messages that are sent are
persisted. However this should be as I think you are saying, until the
sender knows that the sent message has arrived since an acknowledgement
message has been returned.
Does this make sense?
David
-----Original Message-----
From: Stefano POGLIANI [mailto:stefano.pogliani@sun.com]
Sent: Wednesday, January 03, 2001 3:29 PM
To: Martin W Sachs; Burdett, David
Cc: ebxml-transport@lists.ebxml.org; ebxml-tp@lists.ebxml.org
Subject: RE: Yet more issues to resolve
Marty, David,
a little clarification w.r.t. "I also don't like the idea of placing a
restriction on the MSH to persist a message until the transaction is
complete, ..."
I think that an option would be to allow the MSH to be "controlled" by
outside w.r.t. this part and by defining a "default behaviour". "Not
saying
anything" (as mentioned by Marty) would mean, IMHO, that the "default
behaviour" is to remove the message as soon as it reliably reaches the
target MSH.
Giving the option to "control from outside" would allow the layer between
the MSH and the real application to take care (if it will be capable) of a
more sophisticated behaviour.
/Stefano
» -----Original Message-----
» From: Martin W Sachs [mailto:mwsachs@us.ibm.com]
» Sent: Saturday, December 30, 2000 2:23 AM
» To: Burdett, David
» Cc: ebxml-transport@lists.ebxml.org; ebxml-tp@lists.ebxml.org
» Subject: RE: Yet more issues to resolve
»
»
»
» David,
»
» The definition of persistDuration in your email looks OK.
»
» My point about decoupling persistence from reliable messaging was
» a CPA/CPP
» point, not a MSH point. I am simply suggesting that two parties
» might have
» reasons to come to agreement on persistence that have nothing to do with
» the MSH in addition to its use with RM.
»
» Regarding "I also don't like the idea of placing a restriction on the
MSH
» to persist a message until the transaction is complete, ..." I don't
like
» placing that restriction either. I was asking for the opposite - that
the
» MSH specification should not say anything which either prescribes or
» suggests that a message can be deleted from the persistent store
» as soon as
» it reliably reaches the destination MSH. Earlier versions of the RM
» specification included such words.
»
» 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" <david.burdett@commerceone.com> on 12/27/2000 11:20:24
AM
»
» To: Martin W Sachs/Watson/IBM@IBMUS
» cc: ebxml-transport@lists.ebxml.org, ebxml-tp@lists.ebxml.org
» Subject: RE: Yet more issues to resolve
»
»
»
» Marty
»
» In version 0.9a of the spec, a parameter called "pesistDuration"
» is defined
» that is described as follows:
»
» "PersistDuration is the minimum length of time in days that a Message
that
» is sent reliably is kept in Persistent Storage by a MSH. The
» value used for
» PersistDuration is an implementation decision although it MUST be
greater
» than the value of the TimeToLive parameter for any message that is sent.
»
» If a duplicate message (i.e. with the same MessageId) is received before
» the
» PersistDuration has passed, then the MSH that receives it MUST process
it
» as
» a duplicate message as described in sections 10.2.1.1 and 10.2.1.2.
»
» If a duplicate message is received after the PersistDuration has passed,
» then although it may be treated as a duplicate, the sender must realize
» that
» it will probably be treated by the MSH as if the message were a
» new message
» that had not been received before."
»
» So as it stands, the persistDuration only applies to Relilable Messaging
» behavior which I think is right. I know that there are other, e.g.
legal,
» reasons why a messages might be persisted beyond this time. However from
a
» MSH perspective, I think we can only specify the behavior of the MSH and
» nothing else.
»
» I also don't like the idea of placing a restriction on the MSH to
» persist a
» message until the transaction is complete, since the MSH will
» probably need
» to be told by the application that the transaction is complete and it's
» possible that some backend systems will not be able to easily do this.
»
» Do you think the existing definition is OK?
»
» David
»
» -----Original Message-----
» From: Martin W Sachs [mailto:mwsachs@us.ibm.com]
» Sent: Tuesday, December 19, 2000 9:56 AM
» To: Christopher Ferris
» Cc: ebxml-transport@lists.ebxml.org; ebxml-tp@lists.ebxml.org
» Subject: Re: Yet more issues to resolve
»
»
»
» Chris,
»
» Regarding issue 19: I agree that there should be nothing in the TRP
spec
» about how long a message must be persisted since higher layers are
» involved. Perhaps the messaging spec should say that persistence must
be
» at least until the reliable messaging procedure has been
» completed for that
» message transmission. I suggest a non-normative note explaining the
» issues.
»
» In my opinion, at the higher levels, persistence should be at least
until
» the business transaction is concluded. Persistence beyond that
» is probably
» not a matter that has to be agreed to. It is probably an internal
matter
» for each party.
»
» If the persistence element will be in your upcoming CPP/CPA DTD, please
» supply some explanatory text with it. I suggest not coupling the
» persistence element in the CPA to reliable messaging. While it is
» essential to reliable messaging, it is also something that might be
needed
» without ebXML Reliable Messaging. For example, if the parties use a
» transport that is reliable in and of itself, they would probably not
want
» to invoke ebXML Reliable Messaging on top of it but they might want to
» specify something about persistence.
»
» 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
» ******************************************************************
» **********
»
» *********
»
»
»
» Christopher Ferris <chris.ferris@east.sun.com>@east.sun.com on
12/19/2000
» 12:23:17 PM
»
» Sent by: Chris.Ferris@east.sun.com
»
»
» To: ebxml-transport@lists.ebxml.org
» cc:
» Subject: Re: Yet more issues to resolve
»
»
»
» Ian,
»
» Thanks for sending these out. Some comments below.
»
» Cheers,
»
» Chris
» ian.c.jones@bt.com wrote:
» >
» > TR&P people,
» >
» > Sorry this is late - the real world gets in the way
» occasionally.
» >
» > The following issues may now have been overtaken by events but
» we need to
» at
» > least acknowledge this so that I can get the database of issue and
» changes
» > update by the New Year.
» >
» > ####
» > Issue 15
» > Title - RM Info
» > Detail - 7.9.4 - RM Info: why isn't this in routing header?
» >
» > Proposal - Move to Routing Header
» > Resolution -
» > Response By - December 21
»
» I have to disagree with this. The ReliableMessagingInfo
» element applies to the message, not to a particular hop
» along the way. Note that David has crafted a revised section
» on RM in the draft we've been collaborating on which we
» hope to get to the list today (although, we recognise that
» it still needs work).
»
» >
» > ####
» > Issue 16
» > Title - TPAInfo
» > Detail - Add new section containing all TPA(CPA) information
» >
» > Proposal - see above
» > Resolution -
» > Response By -
»
» This one I'm not sure I understand. I thought that we agreed
» in Tokyo to flatten the TPAInfo section by moving all of
» the child elements to be direct decendants of Header.
»
» What information was missing from TPAInfo anyway? I'm not
» clear on what the proposal really suggests.
»
» >
» > ####
» > Issue 17
» > Title - Sequence number
» > Detail - 7.10 SequenceNumber as defined adds little or no value over
the
» > MessageID element required in MessageData. Recommend removing this
» section
» > and the corresponding references to SequenceNumber in later examples,
» the
» > DTD and the schema.
» > If this section remains, must define the exact content of the
» SequenceNumber
» > element when the numeric value is greater than 999. Is the next value
» > "1000" or "1,000"? Is the text for the maximum value "999999999" or
» > "999,999,999" or are both allowed (and equivalent)? Side note:
Numeric
» > formats vary by locale and a comma is not always the thousands
» > If this section remains, "long time" should be defined. At least,
refer
» to
» > a TPA which makes this concrete for a specific trading relationship or
» > mention later text on this subject (if any).
» >
» > Proposal -
» > Resolution -
» > Response By -
»
» Since we're going to spec the schema using XMLSchema, we can
» be more precise in the typing of elements and attributes.
» I agree that the spec needs to be tightened up for this
» particular element.
»
» I'll have to review David's revision of RM to see what that
» has to offer.
»
» >
» > ####
» > Issue 18
» > Title - URI definitions in RM
» > Detail - 7.10 - SenderURI, ReceiverURI and ErrorURI are not
» defined. Are
» > they identifiers or service locations? From later discussion in this
» > document, it appears that ErrorURI is a service location at which
error
» > responses may be received. The other two aren't discussed much.
» However,
» > I'd recommend they be defined similarly to the PartyId elements in the
» > Header (i.e. the parties participating in the current leg of the
» journey).
» > That requires adding a context attribute to SenderURI and
» ReceiverURI and
» > (potentially) renaming the elements. Alternatively, SenderURI and
» > ReceiverURI may somehow identify the service location used in this
» > particular leg (this is probably redundant information for ReceiverURI
» with
» > the transport destination and SenderURI may not exist -- not all
senders
» are
» > reachable in return).
» >
» > Proposal -
» > Resolution -
» > Response By -
»
» I assume that the subject refers to URI elements in RoutingHeader
» not RM. Agree that these need to be more precisely defined.
» As to defining them similarly to the PartyId element...
» I had actually proposed to David in our discussions that
» instead of defining different elements, that we re-use
» To and From but with precise wording that the PartyId
» MUST be an URL form of URI (e.g. something addressable).
»
» This also brings to mind a security issue (possibly) as
» if an intermediate signs a header (with the added RoutingHeader
» element) then it MAY be important to be able to identify
» the intermediary with the specific RoutingHeader. I hadn't
» thought of this before, but it MAY be important to consider.
»
» >
» > ####
» > Issue 19
» > Title - Persistent storage
» > Detail - 7.11 - "Persistent storage" actually lasts how long? Is the
» > original message deleted by the Messaging Service immediately after
» > receiving an acknowledgement? In the case of an error, which
» is reported
» to
» > the sending application (party), does that report also result
» in deletion
» of
» > the original message from the persistent storage?
» > What information does the Receiver store persistently in this step?
» > Recommend storing the MessageID of the original message and the entire
» > response document. To check for "exactly the same as the original
» message"
» > as described on line 828, may be necessary to store entire original
» message
» > or its checksum as well.
» >
» > Proposal - This is out of scope of the TR&P if required it should be
in
» the
» > CPP/CPA
» > Resolution -
» > Response By -
»
» David has added a section on persistence in his RM draft
» proposal. It doesn't get into "how long" as this is correctly
» identified as being something that the parties MUST agree
» upon and have reflected in the CPA. A new element is being added
» to the CPA for this purpose (e.g. to capture the persistenceDuration
» with which a message is to be archived).
»
» >
» > Ian Jones
» > E-Commerce Engineer
» >
» >
»
Powered by
eList eXpress LLC