| Well, what I said below was to default to include, but after walking across the harbor (the hotel buildings here at IETF are really spread out), I convinced myself that _most_ of the users of DUM are likely to assume that it wouldn't be, so I'll change my position. Default to not including it and put red-flashing lights in the doxygen that will catch the eye of someone trying to build automaton/gateways to put it back in. Let me know where that is when you're done and I'll go back in and add a little prose to the comment :)
RjS On Nov 10, 2006, at 8:23 AM, Scott Godin wrote: Just to be clear you think the default should be to include or exclude the Request Uri in the merge detection? I could accept such an option, but it would need to be off by default and heavily documented. The only way this should happen is if the phone _asked_ for two different URIs at some point (by registration or configuration of a peer). It's an acceptable application layer policy to treat those as the same, and the flag you're talking about is exactly stating that policy decision. On Nov 10, 2006, at 6:14 AM, Scott Godin wrote:
I found some older discussions on this same subject: http://www1.ietf.org/mail-archive/web/sip/current/msg10443.html http://www.atm.tut.fi/list-archive/sip/msg06721.html It seems to me that a single UA phone device would want the merge detection as specified in 3261 – ie. why ring for the same call on 2 different lines? But UA’s that handle multiple identities or route calls differently based on request-uri (ie. B2BUA’s, MediaServers, Gateways, etc) would want to include the request uri in the merge detection. I can make this an option in the DUM master profile, if we agree this is a good idea. Scott I went back through 3261 and agree that it doesn't say to include RURI. I have memories of filing this as a bug (and having this conversation on the SIP list several years ago), but I don't find it in the bug tracker now. I'll take it back through the SIP list again. But yeah - we should be On Nov 9, 2006, at 9:18 AM, Scott Godin wrote:
It seems that including the RequestUri in the merged request detection would fix all of these and maintain “actual” merge detection. Is there any reason not to include this (other than it’s not what 3261 says to do)? Scott And this is a problem - this is neither a loop nor a merge - it is two real requests the TU should handle. The redirection is a red-herring from a black-box functionality point of view (if there are artifacts from having sent the redirection to the first request causing the loop we can fix the code so there's not). But this should be handled no differently at the TU level than having the an invite forked by a previous proxy to A and B both of which land at me. If A arrives first and I reject it (with any error response - think 404 instead of 3xx for a moment), then B shows up, my TU needs to handle B as an independent request, not treat it as a merge. (To be complete in the description - if a proxy P1 forked to P2 and P3, and each of those forwarded the request to A at my TU, then the second one arriving I should reject with a 482 - that's the actual merge case). On Nov 9, 2006, at 8:55 AM, Scott Godin wrote:
The scenario you describe with the PSTN gateway boils down to the same merge request detection issue. The problem is possible either with forking or with recursive redirecting proxies. I’ll try to reword Steven’s original description – but I’ll use a PSTN gateway instead of a B2BUA for simplicity. Steps: 1. Call comes in from PSTN to gateway. 2. Gateway passes the call to a recursive redirecting proxy. 3. Proxy recursively redirects to another endpoint on the same gateway. 4. The gateway receives a copy of the invite it sent out, but this time with a different request uri and a loop is detected. Scott Scott - Could you re-sketch the problem for me (an example of the RURI and VIas of both of the arriving requests would help). I'm pretty sure you always want the RURI as part of that check (if someone is running a pstn gateway using resip and a message forks to two different PSTN numbers through that gateway, the second one mustn't get 482ed) , but before we dive into it, lets get a concrete example to work against. On Nov 9, 2006, at 5:54 AM, Scott Godin wrote:
Hmm – that is an interesting problem. Resip/dum is behaving to spec – UA’s are supposed to do merge request detection in this way. But this detection causes problems for B2BUA’s in the scenario you described. Maybe we should have an option for B2BUA’s that allows including the request URI in merge request detection. I’d be interested to hear from others on what they think/know is correct solution to this is. BTW: You mentioned that if you include the request uri in the merge request matching, then you still get a loop detected – I don’t quite understand why that is happening. Scott We have a problem with reSIProcate when operating as a B2BUA between UACs and a proxy IP PBX such as SipX. The scenario is: The standard INVITE call flow is from (1) UAC1 -> (2) B2BUA -> (3) SipX -> (4) B2BUA -> (5) UAC2 Our B2BUA is registrar for UAC1 and UAC2, and the B2BUA also registers to SipX with the ids of UAC1 and UAC2. SipX therefore believes both UAC1 and UAC2 are at the B2BUA … There is only one B2BUA, (2) and (4) represent flows through the B2BUA. This works ok for a normal call. However if at step 4, the INVITE from SIPX to B2BUA is rejected with a 302 to redirect the call to UAC3 (also registered through our B2BUA) , SipX then sends another INVITE to the B2BUA with an updated URI for UAC3 but with the To & From headers unchanged. reSIProcate then returns an error 482 request merged to SipX … there was some commented out code in MergedRequestKey which appeared to check for a change in the request URI but adding this check again did not resolve the issue, 482 was still returned. We are wondering whether the problem is caused because we haven’t cancelled stage (5) before returning the 302 to SipX … Any ideas as to why this happens? I have an ethereal trace available if that helps! Steve Coule Envox Worldwide Envox Worldwide – A Global Leader in Voice Solutions www.envox.com _______________________________________________ resiprocate-devel mailing list
|