http://docs.oasis-open.org/ws-rx/issues/ Date: 2006/10/23 Revision: 2 spec schema soap wsdl policy all New Active Pending Review Deferred Closed Dropped In July comment was made to the WS-Addressing group concerning interoperability issues related to the WS-Addressing WSDL binding and the specific use of the anonymous address specified in the WS-RM public review specification. A WS-Addressing issue cr33 was subsequently opened. It seems clear that after several meetings almost wholly focused on this issue; the WS-Addressing working group is not converging on a resolution. The working group is evenly split between "close with no action" and "keep working toward some resolution". There is a reasonable potential that close with no action may be the result. We therefore request that an issue be raised related to this use of anonymous within the WS-RX Working Group such that alternative forms of reconciliation may be explored. ws-rm spec editorial W3C WS-Addressing WG 1) It's not specified in line 809 that MakeConnection appears as a body element rather than a header. In fact, MakeConnection isn't described as an element at all - it's described as an operation. See line 312 as an example of clearer text. 2) Line 811 says "no SOAP envelopes will be returned on the back-channel." I think "envelopes" should be singular. ws-rm spec editorial W3C WS-Addressing WG Assigned to editors to resolve on Sept. 28th TC call. As chair of WS-Addressing I am forwarding the following comment made by one of our members. I've been puzzling through the spaghetti of dependant specs for a while, and haven't determined conclusively how to reconcile the WSDL in Appendix B with the MakeConnection example in Appendix C.6. The WSDL describes request-response operations such as CreateSequence, with input CreateSequence and output CreateSequenceResponse messages. While the WSDL doesn't describe a binding for this, it is easy to imagine a straightforward way to bind this to a SOAP/HTTP request-response. However, the MakeConnection example shows a MakeConnection message resulting in a CreateSequence response message, which then results in a CreateSequenceResponse messages, followed by an HTTP 202. That is, the first request corresponds to a one-way message (no problem here), the first response corresponds to a request of a request-response, and the second request corresponds to the response of a request-response. What standard binding could be used to describe this behavior? I can't find any of the specs (WSDL 1.1, WSDL 2.0, WS-I BP) that explicitly say the WSDL-described request message must be mapped to an HTTP request, but I'm also not aware of any implementation that allows requests to be mapped to anything else. Is this just a too-obvious-to-state loophole or am I missing something?" ws-rm spec editorial W3C WS-Addressing WG Comments from Gil Comments from Marc ws-rm ws-rmp spec editorial Gilbert Pilz The RM spec says, w.r.t. Seq Ack Ranges: The ranges SHOULD NOT overlap. Why isn't this a MUST NOT? It smells like an interop issue if we don't know if the other side can deal with it. ws-rm spec design Doug Davis Change it to a MUST NOT From spec: Line 893 RM Destinations that generate Sequence faults SHOULD send those faults to the same [destination] as Acknowledgement messages. The text is a bit unclear. Does it mean that sending it is optional or is the EPR it sends it to optional? I think the intent was just the act of sending it is option, but it must go to the AcksTo EPR. ws-rm spec editorial Doug Davis RM Destinations that generate Sequence faults SHOULD transmit those faults. If the fault is transmitted, it MUST be transmitted to the same [destination] as Acknowledgement messages. Stipulate the following scenario: a client and server communicate using asynchronous, two-way, reliable messaging. During the initial sequence creation the client offers a sequence to the server and this offer is accepted. Some number of requests and responses are exchanged using the requested and offered sequences. At some later time the user of the client application causes that application to exit. It is obvious that the client-side RMS should clean up the outgoing, client-to-server sequence by sending a wsrm:TerminateSequence message (possibly preceded by a wsrm:CloseSequence message) to the server-side RMS. What is not clear is what steps the client-side RMD should take to clean up the incoming, server-to-client sequence. There are three basic courses of action: 1.) The client-side RMD could do nothing to clean up the incoming sequence. From a distributed systems viewpoint this seems less than optimal as it leaves the server-side RMS with a "dangling sequence". If the server-side RMS ever tried to use this dangling sequence it would engage the entire RMS machinery for messages that would never get through (send, retry, request acks, retry, request acks, etc.). 2.) The client could wait for a length of time sufficient for it to consider the inbound sequence to be "inactive", then exit. There are a couple of problems with this. Firstly the user that caused the client to exit may grow impatient and attempt to terminate the client by other means. Secondly, outside of the optional expiration time, there is no mutually agreed upon inactivity period for a sequence. 3.) The client-side RMD could send a message to the server-side RMS to indicate that the inbound, server-to-client sequence is being cleaned up. Since the direction of the wsrm:TerminateSequence message is specified by WS-ReliableMessaging 1.1 as being from the RMS to the RMD only, this leaves either the wsrm:SequenceTerminated or wsrm:SequenceClosed faults as the only viable messages for the client-side RMD to use. This brings up all the problems attendant with using faults/exceptions to indicate "normal" application behavior. The server-side RMS may construe the appearance of a fault as an indication of some lower-level problem, etc. Note that, although this issue is discussed in the context of a client application attempting to exit, it is applicable in any situation where the process/VM that "contains" an RMD must exit in a relatively short period of time for whatever reason. ws-rm spec design Gilbert Pilz Amend the wording of section 3.6 (Sequence Termination) to indicate that a wsrm:TerminateSequence/wsrm:TerminateSequenceResponse exchange may be initiated by the RMD. There is interoperability issue with this specicication, since it doesn't define underlying protocol bindings (e.g.,HTTP bindings). The spec or its profile should explicitly define underlying protocol bindings. ws-rm spec design eBusinss Asia Committee 2.1 This specification dose not define how to realize "Duplicate Elimination" and "Message Ordering". It is a serious issue for industry organizations in Japan and Asia to adopt the specification. We strongly recommend to define them to make various implementations interoperable. ws-rm spec design eBusinss Asia Committee 2.2 It is not clear whether AckRequest message can be sent independently or not. The 3.8 describes: "The RM Source MAY request an Acknowledgement Message from the RM Destination at any time by including an AckRequested header block in any message targeted to the RM Destination." In detail, is the following message which doesn't include <wsrm:Sequence> , allowed to send? If so, the above sentence should explicitly describe it. See the first four lines in Section 3.9. <soap:Envelope> <soap:Header> <wsrm:AckRequested> <wsrm:Identifier>http://example.com/test</wsrm:Identifier> </wsrm:AckRequested> </soap:Header> </soap:Envelope> By the way, the section 3.9 describes: "The RM Destination informs the RM Source of successful message receipt using a SequenceAcknowledgement header block. The RM Destination MAY Transmit the SequenceAcknowledgement header block independently or it MAY include the SequenceAcknowledgement header block on any message targeted to the AcksTo EPR." And it is clear the SequenceAcknowledgement can be sent independently. ws-rm spec design eBusinss Asia Committee 2.3 Line697 Editorial "see Section 3.5" should be "see Section 3.8" ? 2.4 Line667 Editorial "see Section 3.1" seems not appropriate. It should be other appropriate section. ws-rm spec editorial eBusinss Asia Committee 2.5 Line 704-709 The following sentences needs to be refined or requires examples. " When the RM Source specifies the WS-Addressing anonymous IRI as the address of the AcksTo EPR, the RM Destination MUST Transmit any SequenceAcknowledgement headers for the created Sequence in a SOAP envelope to be Transmitted on the protocol binding-specific channel. Such a channel is provided by the context of a Received message containing a SOAP envelope that contains a Sequence header block and/or a AckRequested header block for that same Sequence identifier." Maybe some examples help reader understand what it meant. It is not clear whether the SequenceAcknowledgement must be sent on *its* back-channel of the underlying transport protocol. e.g., Here is the following scenario: 1.CreateSequence has "Anonymous" value for AcksTo element. 2.RM Source sends a message A with AckRequested on HTTP Request. 3.RM Source sends a message B on HTTP Request. Question: In this scenario, does RM Destination have to send Sequence Acknowledgement for message A on the HTTP Response in 2, but disallow to send it on HTTP Response in 3? Or does the spec allow to send it on the HTTP Response in 3? The spec is not clear about the above scenario. The spec should clarify it. ws-rm spec design eBusinss Asia Committee 2.6 Line339-348 It is not clear what the following sentences means: CreateSequence\AcksTo = wsa:anonymous Does it mean the SequecenAcknowledgement must be on the HTTP Response to the original message? If so, it is beyond the WS-Addressing definition, I believe. By the way, the spec doesn't allow wsrm:Anonymous to be used in AcksTo. If the spec doesn't allow wsa:anonymous in AcksTo, then it should be explicitly described ws-rm spec design eBusinss Asia Committee 2.7 Section3.9 and Appendix B Needs explanation of ws-addressing composability. It is not clear how ws-addressing/soap request message and response message maps to AckRequested and SequenceAcknoweldgement message. ws-rm spec design eBusinss Asia Committee 2.8 Section3.10 It is not described how you can identify whether the RM destination is polling a message. This could be in RM policy assertion. ws-rm ws-rmp spec design eBusinss Asia Committee The 2nd Protocol Invariant (sec 2.3) says: Within every Acknowledgement Message it issues, the RM Destination MUST include one or more AcknowledgementRange child elements that contain, in their collective ranges, the message number of every message accepted by the RM Destination. The RM Destination MUST exclude, in the AcknowledgementRange elements, the message numbers of any messages it has not accepted. This isn't true in the case when the SequenceAck is either None or Nack. ws-rm spec design Paul Fremantle The purpose of the WSDL is not terribly clear. I doubt one would actually describe a Web service using this WSDL, as RM is generally considered an infrastructure-level protocol and WSDL typically describes an application-level message exchange. In addition, if machine-processing of this WSDL is expected, not all message exchanges are described – after a MakeConnection message comes _in_ to the service, a CreateSequence message may come _out_, yet CreateSequence is only described as an _in_ message in the WSDL. The limits of the WSDL in describing the complete exchanges of messages should be noted. ws-rm spec design Jonathan Marsh fix this editorially by adding the following explanatory text to "Appendix B. WSDL": This WSDL describes the RM protocol from the point of view of an RM Destination. In the case where an endpoint acts both as an RM Destination and an RM Source, note that additional messages may be present in exchanges with that endpoint. Also note that this WSDL is intended to describe the internal structure of the RM protocol, and will not generally to appear in a description of an RM-capable Web service. See RM-Policy [ref] for a higher-level mechanism to indicate that RM is engaged. 1. Section 3.2 (P. 10) This section describes piggy-back messages, but does not mention receivable message combinations. So, which message combinations can be received could be implementation-dependent. ws-rm spec design Sadao Yashiro 2. Section 3.8 (P. 19) According to the specification, a receiver must return an Acknowledgement when a message contains an AckRequested element, which requests an Acknowledgement. However, it is not specified that a receiver must not return an Acknowledgement when a message contains no AckRequested element. So, I think the implementation should return an AckRequested element whenever a user payload is received. ws-rm spec design Sadao Yashiro 3. No description on guaranteed message ordering and duplication elimination. Neither did the previous version. Did the TC determine to set them out of scope? ws-rm spec design Sadao Yashiro WS-RM allows for the possibility of adding the wsrm:AckRequested and wsrm:SequenceAcknowledgement headers to unrelated messages that are targeted to the "same" endpoint; a concept it refers to as "piggy-backing". There are a number of problems with this concept: See message for complete description. ws-rm spec design Gilbert Pilz WS-RM 1.0 [1] defined a sub-element of the wsrm:Sequence header that served to mark the containing message as the last message in a Sequence. WS-RM 1.1 [2] has removed this element. Consequently the RMD has no guaranteed way of determining whether it has received all the messages in a Sequence. This presents obvious drawbacks to an Application Destination that may wish to know if it has received all the data that the Application Source sent it. In addition to this it makes correctly implementing the "incomplete sequence behavior" semantics impossible since the RMD cannot always determine what is and isn't an "incomplete Sequence". For example, suppose an RMS creates a Sequence, sends messages 1-10, then sends a CloseSequence. Suppose that messages 9 and 10 get lost but the CloseSequence message is received by the RMD. The RMS can determine that the Sequence is incomplete (final ack is missing 9 and 10 from the range), but the RMD has no way of knowing, nor is there any way for it to discover, that the Sequence is incomplete. If the IncompleteSequenceBehavior was "DiscardEntireSequence" then the RMS will conclude that all of the messages will be discarded whereas the RMD will, in all likelihood, deliver all of the messages to the Application Destination under the assumption that the Sequence is complete. ws-rm spec design Gilbert Pilz In section 4.6 the CreaetSequenceRefused fault is defined as a sender fault. This fault should apply to sender or receiver. ws-rm spec design Marc Goodner In section 4.7 describing the SequenceClosed fault one reason the fault must be returned is that an RMD is “…asked to close a Sequence that is already closed.” However, in section 3.5 that describes CloseSequence it says that for a closed sequence the RMD must process all Sequence Lifecycle messages including CloseSequence. It also says that subsequent CloseSequence messages have no further effect. ws-rm spec design Marc Goodner