[sr-dev] Outbound registrar.lookup() not ordering results by reg-id?

[Olle E. Johansson]

4 mar 2013 kl. 17:57 skrev Daniel-Constantin Mierla <miconda at gmail.com>:

> 
> On 3/4/13 5:35 PM, Olle E. Johansson wrote:
>> 4 mar 2013 kl. 17:06 skrev Daniel-Constantin Mierla <miconda at gmail.com>:
>> 
>>> Hello,
>>> 
>>> On 3/4/13 11:55 AM, Peter Dunkley wrote:
>>>> Hi,
>>>> 
>>>> I've been looking at caching the ruid and while reading the code I have not been able to understand how the contacts get correctly ordered for outbound.
>>>> 
>>>> The contacts should be ordered so that those with the lowest reg-id are the first tried for each ;+sip.instance.  However, I can't see any comparisons of reg-id anywhere (I've looked in usrloc, registrar, and the tm:t_..._contacts() functions).
>>>> 
>>>> Have I missed something or is this (like the parallel forking issue Olle reported from SIPit) another outbound registrar issue that we need to fix?
>>> location module keeps the records ordered by registration time, iirc. By default, parallel forking does not care of Q value, will do branching at once to all contacts.
>>> 
>>> If you need to take in consideration Q, then you have to use t_load/next_contacts(). I assume it is the same for outbound and reg-id, Juha implemented both of them.
>>> 
>>> In case the issues is something else, please refresh me with a link in archive (or add to tracker) about the details after SIPit, I guess I missed some discussions due to heavy traveling and may take me a while to spot it in the history.
>> Here's the issue:
>> 
>> Like with the example of a 302 for one call leg in a parallell fork, we need to handle one call leg here too.
>> 
>> Let's assume I have a SIP account miconda at edvina.net and you register two phones with outbound, which means at least four contacts, in groups of two sharing the same +sip.instance but have different reg-id's.
>> 
>> If I call these and one phone has an issue, then the edge proxy reports 430 flow failed and I need to move to the other reg-id contact for that phone only.
>> 
>> As far as I understand today, I can't retarget unless I hit failure_route. I won't hit that as long as our other phone is ringing.
>> 
>> That's why I suggested we consider a failure_branch_route() for these kind of retargets that won't affect the fork. This is like the earlier example on the mailing list where I forked to multiple phones, one sent 302. I would like to follow that 302 before the fork completes and ends up in failure_route or gets a 200 OK.
> I think this is quite a debatable aspect, which is more a handling preference than a issue, so eventually can end up in some config option.
> 
> First, a 302 should be normally sent back to caller -- that only after all branches were processes, otherwise the caller has no idea there are pending active branches on initial try and destroy dialog in its side. A 3xx received by caller usually creates completely new dialogs. Handling 302 on proxy is admin decision, not a SIP specs proxy requirements.
> 
> Personally I would not like to follow blindly a 3xx unless all branches are completed with no answer, which differs from your preference, thus again the need of an eventual configuration option (mod param, function, event_route, etc...). But I don't find this a bug at all, it's been like this from ser 0.8, I guess.
> 
> Regarding the flows priorities, I think the logic should be in tm (in t_*_contacts*() or whatsoever), as said in previous email. usrloc/registrar just populates destination set, it is tm that can do serial/parallel forking. Even if not using t_*_contacts*(), that means parallel forking to all records (no matter of flows), which eventually cannot end up in something very bad, call being answered on first picked up phone. Again, like with Q in normal contacts, a matter of admin willing to take it in consideration or not.
> 
> From my point of view, this logic of selecting flows by priority should be in those tm functions added by Juha (to follow the same architecture as with Q), if not already there can be added at any time. But I don't find it really a blocker.

I agree it's not a blocker, but there's another point I want to make here too.

I've always been trying to be strict in marketing - don't promise too much and try to be honest. If we claim to support a feature, we should support it. I don't like people saying that Asterisk 11 supports WebRTC, since it doesn't even support the current proposals, and the technology is still under development. Asterisk 11 will likely never support WebRTC because that would violate the current release policy.

Now, if a core feature in outbound is missing we can not claim that Kamailio 4.0.0. supports outbound. Failover between flows in outbound is a core feature of the outbound standard. We can say that we have an early development release of an outbound module and fix it for a later release.  Sometimes things just doesn't finish on time in Open Source, where we depend on contributions of individual time or commercial funding. I think we have to communicate clearly here and tell our user base that this module is not finished.

Regarding the handling of 302 - yes, I agree with you in theory - the RFC claims what you say. Unfortunately not all implementations agree with the standard. There are many DND buttons on SIP phones that send a 302 that you want to handle in the core network. The DND could lead to a secretary and her phone should start to ring while other phones are still ringing. Every case is different and has to be handled in the configuration script on a case by case basis. 

For a softphone 302 where companies seldom has control over the configuration, I wouldn't trust the 302 in the core, so I would send it back all the way if needed.

Cheers,
/O