- From: Doug Davis <dug@us.ibm.com>
- Date: Tue, 10 Oct 2006 13:08:33 -0400
- To: public-ws-addressing@w3.org
- Message-ID: <OF5C59B726.A03ABAFF-ON85257203.005A1FA4-85257203.005E29E9@us.ibm.com>
Alastair, There's a component of WSA that people are not really taking to account when thinking about solutions to this problem, and that's the EPR itself. One of the many good things that WSA has given us is that standard/consistent/reusable structure of how to refer to an endpoint. Whether we want to share this information with another soap node, so we put it into a soap envelope some place, or whether we are passing this information around inside of our soap stacks, WSA has provided a very nice wrapper for this information. In the past we may have just passed around a URI/String, but the EPR is far more flexible and useful than that. I believe that once people have adopted WSA the switch from URI/String to EPR is one that is very wide-spread and nontrivial. Once it is done most of the code that may interact with an EPR really doesn't need to do anything with the data inside of the EPR itself. In most cases the only time you would ever really need to look inside of it is at the time the soap stack actually tries to send a message to that EPR. And even then, I would hope its just the wsa:Address value that is examined - the other bits (ref-p's) are just blindly echoed as headers. So, if people are passing around EPRs I get really worried at the suggestion that something else needs to be passed around as well in order to correctly talk to that EPR. The thought that people will need to revamp all of their code that deals with EPRs to also pass along additional information is something that should not be taken lightly. I would also suggest that whether we are pushing messages that EPR (async) or whether we are pulling messages for that EPR (makeconnection), the logic above the transport layer should be blissfully unaware of it. While as with all good rules there are exceptions, in the general case I see no reason why the application should concern itself with whether the wsa:Address field of the [destination] EPR is http://www.cnn.com, http://.../wsa/anonymous, http://..../wsa/none, or http://.../wsrx/anon. The specific value of the URI is really not of any concern to it. This abstraction that the EPR construct provides is a very powerful and useful one - and we should strive to maintain it. To your suggestion that there can be additional headers - given all of the above, I do not believe any of that information should be external to the EPR. Correlation related to semantics above the transport layer (e.g. what Tx are we using) are things that make sense as additional headers but information directly related to the endpoint reference should be part of the EPR. And I have similar concerns with the wsa:RelatesTo proposal. These proposals would, IMO, break existing code that assumes an EPR is all that is needed when passing around references. From personal experience (from a coder's perspective) the impact of having to pass around anything more than what we already do (meaning an EPR) would be huge (not to mention, IMO, the wrong choice :-). thanks -Doug Alastair Green <alastair.green@choreology.com> 10/09/2006 08:40 PM To Christopher B Ferris/Waltham/IBM@IBMUS cc Marc Hadley <Marc.Hadley@Sun.COM>, Doug Davis/Raleigh/IBM@IBMUS, public-ws-addressing@w3.org, public-ws-addressing-request@w3.org Subject Re: An Example Use of RM's MakeConnection I think we need to take a step back. The example I gave (shorn of XML) is in my view exhaustive. If you can support this use case, then you're done -- for all applications (including RM) that need to create a bilaterally-initiated conversation out of a monologue. Here it is again: Given a connection identified as X, between endpoint A (that is non-addressable) and endpoint B (that is addressable), I posited the following canonical use case in a recent posting to WS-RX: A req: message 1, connection X, reply to anon if available. B resp: message 2 A req: message 3, connection X, reply to anon if available B resp: no envelope, 202, i.e. message not available A req: MakeConnection, X B resp: message 4, MessagePending A req: MakeConnection, X B resp: message 5 Net effect, for connection/conversation identified as X: A to B: message 1 B to A: message 2 A to B: message 3 B to A: message 4 B to A: message 5 >From this you can draw out four semantic elements for which someone must define types and syntax: 1. An identifier for the conversation (X is an instance of this type) 2. An indication that a response on the back channel is legitimate (but optional), if the recipient has a message that belongs to the conversation X 3. A no-op message, that invites an (optional) message on the back channel that belongs to the conversation X 4. An indication that it is worth sending no-op message 3 again, because the number of responses to a message bearing indication 2, or a message of type 3, is > 1. These are elements of a protocol that is a) nothing inherently to do with WS-RM (one can imagine other uses) and b) nothing inherently to do with WS-Addressing as it currently stands (you can do it all without WS-A as it is today, hence the impulse to waive WS-A 3.4.) Examples of things like X are: conversation identifiers, and sub-conversation identifiers, which map to a conversation identifier (UUID and sequence id, in RM's case). Concretely, for a protocol called TalkToMe, ns prefix T: 1) A header element <T:Identifier/> whose content is any application-meaningful element (e.g. a UUID, or a sequence identifier), and whose presence imports: "know this identifier, and respond to it on the back channel of this or any subsequent message containing the same identifier" 2) A body element <T:Pull/>, which must have present and associated a header <T:Identifier>, whose back-channel response may be any message relating to the value of <T:Identifier>. 3) A header element <T:MessagePending/>, which should only be present in a message on the back-channel which relates to a prior message carrying the header element <T:Identifier/>, which imports: "I have another message relating to this conversation to send you". If WS-Addressing defined this protocol, then we could all use it, and all users would stop pestering WS-Addressing. In other words, the "novel, unorthodox, interesting" MEP would find a natural home, and become "customary, acknowledged, orthodox, usual -- and generic". Each application would then be responsible for defining its valid types for the child element (content) of <T:Identifier/>. This approach fully exploits the self-correlating nature of the back-channel. I can't see why we need to explicitly state a relationship that is unambiguous by its very fact. The identifier needed does not relate to an individual message, but to a conversation. Alastair Christopher B Ferris wrote: Marc, Fine for handling request/response I suppose. However, not everything is request/response. How does this work when there is no "response" but rather a need to get messages from the "server" to the "client"? Cheers, Christopher Ferris STSM, Software Group Standards Strategy email: chrisfer@us.ibm.com blog: http://www.ibm.com/developerworks/blogs/page/chrisferris phone: +1 508 377 9295 public-ws-addressing-request@w3.org wrote on 10/09/2006 12:26:07 PM: > Looking at the message flow, I think WS-RM could make more use of > wsa:RelatesTo instead of inventing a new anon URI. Here's the same > message flow using the WS-A anon URI and making more use of > wsa:RelatesTo and @RelationshipType. > > I think this formulation removes the requirement for additional WSRM- > specific anon URIs, let me know if I missed anything. > > Marc. > > Scenario: Client sends GetQuote to server unreliably. Server wants > to send GetQuoteResponse using RM so it must sent a CreateSequence > before I can send the GetQuoteResponse back. > > Step 1 - Client sends GetQuote to Server > <soap:Envelope ...> > <soap:Header> > <wsa:To> http://stockquote.com </wsa:To> > <wsa:Action> foo:GetQuote </wsa:Action> > <wsa:MessageID> uuid://.../100 </wsa:MessageID> > <wsa:ReplyTo> > <wsa:Address>http://www.w3.org/.../anonymous</wsa:Address> > </wsa:ReplyTo> > </soap:Header> > <soap:Body> > <foo:GetQuote> IBM </foo:GetQuote> > </soap:Body> > </soap:Envelope> > > Step 2 - Server sends an RM CreateSequence to the Client using the > only means it has available - the transport backchannel. > <soap:Envelope ...> > <soap:Header> > <wsa:To>http://www.w3.org/.../anonymous</wsa:To> > <wsa:Action> http://...wsrm/CreateSequence </wsa:Action> > <wsa:MessageID> uuid://.../101 </wsa:MessageID> > <wsa:RelatesTo RelationshipType="http://...wsrm/InitReliable"> > uuid://.../100 > </wsa:RelatesTo> > <wsa:ReplyTo> > <wsa:Address> http://stockquote.com </wsa:Address> > </wsa:ReplyTo> > </soap:Header> > <soap:Body> > <wsrm:CreateSequence> ... </wsrm:CreateSequence> > </soap:Body> > </soap:Envelope> > > Notice the use of RelatesTo with a WSRM-specific @RelationshipType to > indicate that this message is a WSRM-specific response to the initial > request. > > Step 3 - Client sends a CreateSequenceResponse to wsa:ReplyTo > <soap:Envelope ...> > <soap:Header> > <wsa:To> http://stockquote.com </wsa:To> > <wsa:Action> http://...wsrm/CreateSequenceResponse </wsa:Action> > <wsa:RelatesTo> uuid://.../101 </wsa:RelatesTo> > </soap:Header> > <soap:Body> > <wsrm:CreateSequenceResponse> ... </wsrm:CreateSequenceResponse> > </soap:Body> > </soap:Envelope> > > Step 4 - Having not received the GetQuoteResponse, the Client uses > MakeConnection to allow it to flow back > <soap:Envelope ...> > <soap:Header> > <wsa:To> http://stockquote.com </wsa:To> > <wsa:Action> http://...wsrm/MakeConnection </wsa:Action> > <wsa:ReplyTo> > <wsa:Address>http://www.w3.org/.../anonymous</wsa:Address> > </wsa:ReplyTo> > <wsa:RelatesTo RelationshipType="http://...wsrm/InitialRequest"> > uuid://.../100 > </wsa:RelatesTo> > </soap:Header> > <soap:Body> > <wsrm:MakeConnection> > </wsrm:MakeConnection> > </soap:Body> > </soap:Envelope> > > Notice the use of the wsa:RelatesTo with a WSRM-specific > @RelationshipType to indicate that this message is requesting a > response to the initial request message. > > Step 5 - Server uses the backchannel to let the GetQuoteResponse flow > back to the Client > <soap:Envelope ...> > <soap:Header> > <wsa:To>http://www.w3.org/.../anonymous</wsa:To> > <wsa:Action> foo://GetQuoteResponse </wsa:Action> > <wsa:RelatesTo> uuid://.../100 </wsa:RelatesTo> > <wsrm:Sequence> ... </wsrm:Sequence> > </soap:Header> > <soap:Body> > <foo:GetQuoteResponse> 139.0 </foo:GetQuoteResponse> > </soap:Body> > </soap:Envelope> > > Notice the wsa:RelatesTo points to the GetQuote request message and > it is sent using RM (the Sequence header), and that the SOAP Envelope > looks exactly like it would if it had been sent on the original > transport backchannel - meaning, the wsa:To is derived from the > wsa:ReplyTo from the GetQuote request message not the MakeConnection. > > --- > Marc Hadley <marc.hadley at sun.com> > Business Alliances, CTO Office, Sun Microsystems. > >
Received on Tuesday, 10 October 2006 17:08:56 UTC