- From: Mark Nottingham <mnot@mnot.net>
- Date: Fri, 27 Apr 2001 11:50:13 -0400 (EDT)
- To: Roger Wolter <rwolter@microsoft.com>
- Cc: Mark Nottingham <mnot@akamai.com>, Frystyk <frystyk@microsoft.com>, xml-dist-app@w3.org
This sounds like a higher-level design issue for the application (or the modules that it chooses to use ;). There are a number of conditions that may prevent the client from receiving or understanding the server's response, without the server knowing this. A well-designed ordering system might have an order rollback mechanism, and/or an order verification mechanism. On Fri, Apr 27, 2001 at 08:12:05AM -0700, Roger Wolter wrote: > I think this is especially dangerous in a RPC style communication. For > example, if a client sends an order for 1000 cases of blue widgets to a > server, the server processes the order and returns an acknowledgment > with a mustUnderstand header that the client can handle, do we really > want the Soap processor on the client to generate an error? What > happens to all those widgets? How does the client determine whether the > order succeeded or not? May be better approach would be to return a > warning that made it clear that the action succeeded but processing the > response failed. > > -----Original Message----- > From: Mark Nottingham [mailto:mnot@akamai.com] > Sent: Monday, April 23, 2001 12:55 PM > To: Frystyk > Cc: soapbuilders@yahoogroups.com; xml-dist-app@w3.org > Subject: Re: mustUnderstand on client side > > > > Just thinking out loud... > > There seem to be a number of conflicting interests re: what do to > with a fault. It seems *most* tied to an application's message > exchange pattern (i.e., request/response will probably send a fault > back in the response, one way will probably propogate the fault to > the destination if possible, etc.). I can imagine that default fault > handling behaviours might be defined for each, and a module can be > built that will explicitly tell processors what do do with faults in > the exception cases. > > > > On Sun, Apr 22, 2001 at 07:11:57PM -0700, Henrik Frystyk Nielsen wrote: > > > > The basic SOAP processing model described in [1] doesn't know about > > servers or clients but only "SOAP processors" which is the reason why > > SOAP in general walks a fine line when talking about *generating* a > > fault vs. *sending* a fault. The latter is only mentioned in the HTTP > > binding and the RPC convention as these two talk about "requests" and > > "responses". That is, SOAP distinguishes between generating a fault > and > > sending a fault message somewhere. > > > > Within the context of a request/response model, I think the right > thing > > for an HTTP client that cannot accept/obey the SOAP message semantics > is > > to fault the processing or to suggest that the message is saved for > > later processing by a more savvy processor. The latter is similar to > > what clients do when receiving a response with an unknown media type. > > > > It is important that a SOAP processor doesn't break the processing > model > > and merely ignores the SOAP rules just because it may not be in a > > position where it can notify the sender about the fault. > > > > Henrik > > > > [1] http://www.w3.org/TR/SOAP/#_Toc478383491 > > > > Doug Davis wrote: > > > > > >I see 3 choices: > > > 1 - ignore it > > > 2 - fault back to the server > > > 3 - fault back to the client > > > > > >While 2 is probably what "should" happen, I doubt many > > >will be able to support this. 1 seems most likely, but > > >scares me. 3 seems like a nice middle of the road solution. > > >8-) > > > > > >Paul Kulchenko wrote: > > > > > >Since it's possible to return Header from server to client > > >also, what should client do if this header has mustUnderstand > > >attribute or wrong actor? > > -- > Mark Nottingham, Research Scientist > Akamai Technologies (San Mateo, CA USA) > -- Mark Nottingham http://www.mnot.net/
Received on Friday, 27 April 2001 20:09:45 UTC