Pim:
as discussed today, we have two solutions to pick from,
none of them being a clear win...
Let me just try to alleviate some of the drawbacks you see
to WS-addressing:
From: Pim van der Eijk
[mailto:]
Sent: Wednesday, May 28, 2008 2:02
PM
To: 'Moberg Dale';
Subject: RE: [ebxml-msg] MOU on
intermediate support by conformance profile and other auxiliary documents for
ebMS 3.0
Here are some reasons why I would
prefer to use ebMS headers for routing rather than WS-Addressing Reference
Parameters:
Conformance to v3.0 Core
The ebMS 3.0 Part 1 Core does not specify the use of WS-Addressing.
<JD> Right, although we know the use of RM (WS-RX)
automatically requires some level of support to wsa. RM Acks need be routed
based on the content of wsrm:CS/ AcksTo, which is an EPR. It will be expected
this EPR has enough value to either directly send the Acks, or to route them.
The associated ref parameters are supposed to be put in the header and are a
natural fit for routing... All this was not so crucial for point-to-point
exchanges where an "anonymous" value is expected most of the times in the
AcksTo
Our
engineering believes it is easier to do an "wsa extension" to a Core MSH
vs. to add generating a (dummy) eb:header in some cases and is a bit cleaner
from architecture viewpoint.
Avoid yet another standard to
support
Requiring support for WS-Addressing in a multihop context adds a burden
to implement ebMS 3.0 in general and the multihop profile in particular.It is
also true that the counterproposal requires adding ebMS headers in some
situations where v3.0 Core does not require them, but any ebMS 3.0
implementation needs to know how to construct or handle ebMS 3.0 headers anyway,
so this is just a reuse of that general functionality.
<JD> WS-addressing support is greatly facilitated by current
WS stacks which provide APIs to add / consume wsa headers.
We
need to look at the corner cases as well, where I think wsa may have an
edge:
- if an
ebMS intermediary has to generate an eb:Error, the routing of this error cannot
always rely on a "reverse reading" of the eb:Header of the faulty message for
back routing. Wsa provides wsa:From and wsa:ReplyTo headers that contain the
same value type (EPR) as the wsrm:AcksTo, and that could be used for
this.
- we
still have the problem of routing eb:Signals (Receipts, Errors) that originate
from the ultimate receiver. Although we can generally assume the eb:header of
the related User message can be used for reverse-routing of these, I feel more
comfortable with an alternate capability that allows for different routing data
including a direct URL when possible (in same way as wsrm:AcksTo does for RM
Acks). wsa:ReplyTo and wsa:From support
this.
- what if
some User messages have new header data that are intended to the same MSH as
others, but for which I-Cloud routing functions have not yet been updated (e.g.
a business partner behind a MSH decides that its sub-departments will now have
their own PartyId). I like the idea that one can still reuse an old EPR data
that is known to route properly, assuming that when both wsa ref param headers
AND eb:header are present, routers look first at wsa ref param
headers.
-
bundling: as discussed in F2F, it could be that several UserMessage
elements have to be bundled. Relying on eb:headers for routing in such cases,
imposes several restrictions on how bundling can be done. Giving precedence to
wsa ref param headers for routing is an unambiguous solution (see previous
case of not up-to-date routing
functions).
Inconsistency
If we start using WS-Addressing reference parameters in the multihop
context, the question comes up why we did not use them in ebMS 3.0 Core. There
are already elements in the ebMS header that are also in WS-Addressing, such as
MessageId. Nevertheless, v3.0 Core does not use WS-Addressing even for these
header fields, let alone use reference parameters for the information types that
are in an ebMS 3.0 header but are not regular WS-Addressing field. I think the
ebMS 3.0 header is richer than WS-Addressing header, so that this decision was
right at the time. But even if it were not, I think a multihop profile should
stick to the approach taken in v3.0 Core.
<JD> we discussed a lot about wsa:MessageId in the past. One
of the reasons to not rely on it was in case of bundling: the identity has to be
per message unit (eb:UserMessage or eb:SignalMessage). We already have in Core
V3 some limited cases of bundling Signal + User units. eb:Errors may need to
selectively refer to these based on
eb:messageId.
Limited reuse
It might look as if using WS-Addressing is an instance
of reusing a more general specification instead of defining ebMS specific
mechanisms. But unlike the reuse of WS-Security or
WS-Reliability/WS-ReliableMessaging, this reuse is only a syntactic issue. We
would just be adding information to some defined extension points in the schema.
Any processing of that information would require ebMS-specific processing,
rather than using generic WS-Addressing functionality. Any intermediary that
provides WS-Addressing functionality and wants to conform to this profile would
still need to have ebMS-specific functionality.
<JD> But we could say as well
that it is in the nature of wsa to be a "syntactic specification", i.e. unlike
WS-RM or WSS it does not define precise mechanisms: only some mappings between
EPR construct and header content, as well as some limited control of the HTTP
binding and some cross-message correlation features. Now, besides the
header aspect, the notion of EPR appears to be a convenient modeling construct that can be reused
in PModes, and is able to wrap up all data needed for routing, with ref
parameters.
Duplication in message
structure
Having to store or retrieve the same type of information in two different
locations usually smells of missed generalizations, and I think that is true
here too. We looked at putting the equivalent of an ebMS header in a URL (the
earlier HTTP URL query parameters), I don’t find putting it in WS-Addressing
parameters a lot more elegant. It also raises issue like how to handle any
inconsistencies in situations where there is both a WS-Addressing structure and
an ebMS header. Also, the WS-Addressing specification says that using reference
parameters to store information that would otherwise be in other SOAP headers
could
cause “unintended or erroneous semantics in the resultant SOAP message".
<JD> this is a possibility. But if we make it clear what the
routing precedence rules are, we have to consider the good side of
discrepancies: in some cases - e.g. delay in updating the routing
functions: you don't want to or can't rely on the eb:header. So discrepancy may
be intentional. Given taht the eb:header is primarily intended for message
binding after reception, while wsa ref params are only for routing, having an
exact map should not be always expected or
required.
Generalized routing
This is related to duplication. When using just ebMS
headers, assuming routing is done
on destination PartyId, any
message (user, signal, WS-* helper, bundles, piggy backs) can be routed based on
the single following XPath expression:
S:Envelope/S:Header/eb3:Messaging/eb3:UserMessage[1]/eb3:PartyInfo/eb3:To/eb3:PartyId
<JD> assuming the use of dummy eb:headers, yes. But again,
there are seveal other factors to consider.
From: Moberg Dale [mailto:]
Sent: 07 May 2008 23:14
To: Pim van der Eijk;
Subject: RE: [ebxml-msg] MOU on
intermediate support by conformance profile and other auxiliary documents for
ebMS 3.0
Yes, I agee that adding
metadata to ebMS signals or to other messages used in WS protocols is what
is agreed upon as a MOU item.
The way or ways to do
this is apparently TBD J
I think that since we
have several ways to attach the metadata, we probably want to prune these down
into a simple convention to do it in a standard
way.
(Incidentally, would we
then provide CPPA support for these messages? Would they be treated as msh level
messages from the default packaging and delivery channel
standpoint?)
From: Pim van
der Eijk [mailto:]
Sent: Wednesday, May 07, 2008 1:04
PM
To: Moberg Dale;
Subject: RE: [ebxml-msg] MOU on
intermediate support by conformance profile and other auxiliary documents for
ebMS 3.0
We agreed to add ebMS
metadata also to signals and helper messages. And we also looked in some
detail at one option for adding this metadata, by adding it to a WSA
header. Imo we did not fully compare this option to using a an
eb:Messaging header (containing an eb:UserMessage but one that has
no eb:PayloadInfo) to convey the ebMS metadata. I want to fully
understand both options and compare their pros and
cons. Unfortunately, due to travel I will not be able to attend next
week's meeting.
Pim
From: Moberg
Dale [mailto:]
Sent: 07 May 2008 19:42
To:
Subject: [ebxml-msg] MOU on intermediate
support by conformance profile and other auxiliary documents for ebMS
3.0
Goals
Transparency where feasible
Signature intact where feasible (important aspect of
transparency)
WSI Conformance where feasible (especially with respect
to WSI RSP policy that WS-RX Reliable Messaging headers be signed, including
WS-Addressing elements where used within
ReliableMessaging)
Spoke configuration to services within an I-Cloud should
have simplicity of configuration change management.
Specifically, rerouting to services (within the I-Cloud
) can be accomplished without spoke configuration
changes”)
Hub service destinations and spoke clients have
“vanilla” ebMS 3 conformance where feasible. (Some specializations of general
functionality and of extension points is allowed.)
Additional
Implicit Constraints
The internal addresses within an I-Cloud can have
private IP addresses and DNS names that are not publicly reachable or resolvable
in the Internet.
Proposed
Rerouting by intermediaries can be based on ebMS
“metadata” both in forward and return paths.
This functionality is referred to as a “table” mapping
ebMS metadata to next hop URLs so that the HTTP POST (or other?) command can be
rewritten.
TBD: This map may be augmented for return
paths by keeping a map involving Message-Id and Relates-To values.
For ebMS user messages, no special treatment is needed
(so far anyway).
For ebMS signal messages and for WS clerical messages
(such as CreateSequence, CreateSequenceResponse, etc.) several approaches are
under consideration
1. Define a ebMS Reference Parameter that allows
the wsa attribute “isReferenceParameter” applied and that contains the metadata
needed for routing.
2. Allow use of WS-Addressing headers such as
From or ReplyTo or FaultTo that have an EndpointReference model (and include
ReferenceParameter within the EPR).
3. For return path, if WS-Addressing Messaging-Id
was present in incoming message, require use of RelatesTo in response
message.
4. For using the request connection for a
response, some read-only access to ReplyTo is needed to check for the
“anonymous” URL value.
In this case, all
intermediaries would be expected to hold the connection open (?) to allow the
final service to use the HTTP backchannel for its
response.
For Pull MEP-binding, allow first intermediary to handle
MPC requests and let internal I-Cloud arrangements for this option be
implementation specific.
