[OpenSIPS-Users] Two Opensips proxies sharing the same DB

[Jeff Pyle]

James,

I think you're going to have to run usrloc in dbmode=3, and then use a
manual MYSQL() call in Asterisk to pick it off the location table so you
know which Opensips instance to send it to.

I don't have a lot of experience with NAT'd clients.  Will a client
register to the same place each time?  If not, is there a chance it could
change mid-call, and if the call were long enough, you'd lose your pinhole
from the proxy handling the call back to the client.


- Jeff


On 1/29/11 4:13 PM, "James Lamanna" <jlamanna at gmail.com> wrote:

>Hi Jeff & Bogdan,
>I'm looking into a setup very similar to this as well, essentially I
>want to have a cluster of OpenSIPS servers for registration and then a
>cluster of Asterisk Boxes for all the dialplan handling.
>I have the unfortunate problem that all of my clients are going to be
>behind NAT.
>My main issue I think is going to be dealing with NAT, and I'm looking
>for any ideas here.
>
>My initial idea is this:
>
>- Identify an OpenSIPS server as a "relay" between Asterisk Servers
>and the OpenSIPS Servers.
>- On an incoming call from PSTN/Asterisk, divert the call to the relay
>to lookup which OpenSIPS server the callee is registered on.
>- Redirect the INVITE to this OpenSIPS server, so that the call can be
>send back appropriately through NAT.
>
>That seems reasonable, however, what's the best way to lookup the
>socket column to determine which server the callee is registed on? (or
>is there another way)
>
>Thanks.
>
>-- James
>
>
>2011/1/20 Jeff Pyle <jpyle at fidelityvoice.com>:
>> Bogdan,
>>
>> I think I've got it now.  Details inline.
>>
>> On 1/20/11 3:44 PM, "Bogdan-Andrei Iancu" <bogdan at opensips.org> wrote:
>>
>>>Jeff Pyle wrote:
>>>>>> We're looking to add a second Opensips instance on a separate server
>>>>>> for failover. Or, from an operational perspective, it could be
>>>>>> described as "active-active" since both will be available at any one
>>>>>> time. We'll control the traffic flow to the proxies with the SRV
>>>>>> records used by the clients.
>>>>>>
>>>>>> Looking through the db tables used, it seems there may be some
>>>>>> conflict with the location and dialog tables.
>>>>>>
>>>>>> The usrloc module clearly saves the local socket used during the
>>>>>> registration. Is there a way to tell Opensips 1.6 to ignore this
>>>>>>when
>>>>>> loading the record?
>>>>>>
>>>>> socket;s are discarded at load time if not local.
>>>>>
>>>>
>>>> I run db_mode=3 to keep everything current in the db.  The performance
>>>>is
>>>> acceptable (MySQL cluster helps).  If I understand you correctly:  on
>>>>a
>>>> shared table, if I save() a registration on Proxy A, but then load()
>>>>it
>>>> into Proxy B, Proxy B will ignore it since the socket is non-local.
>>>the socket is ignored, not the record - the record will be used, but the
>>>socket info discarded
>>
>> Ah ha!  So it will work on a shared table, and the default socket of
>> "Proxy B" will be used in the scenario I described. Excellent.
>>
>>>
>>>>   Any
>>>> way to work around that within Opensips itself?  If I ran separate
>>>> location tables on each, I might be able to work something up with
>>>>MySQL
>>>> triggers to push a saved registration from one table to the other at
>>>> save()-time, changing the socket field as it goes.  That's a bit more
>>>> hackery than I was hoping to have to implement.
>>>>
>>>>
>>>>>> That way, either proxy can use records saved by either proxy. A
>>>>>>force
>>>>>> socket option perhaps to the local IP? Clients are all public IPs ­
>>>>>>no
>>>>>> NAT.
>>>>>>
>>>>> depends on what db_mode you use for usrloc - if you use a DB mode
>>>>> involving caching, the DB is read only at startup (otherwise, at
>>>>> runtime, it is just written), so data will not be shared at all. Of
>>>>> course you completely disable the usrloc caching via DB_ONLY db_mode,
>>>>> but the performance penalty is high - maybe you should consider
>>>>>register
>>>>> replication at SIP level.
>>>>>
>>>>
>>>> Replication at the SIP level.  That is, have the SIP client register
>>>>with
>>>> both proxies at once?
>>>>
>>>yes. UAC registers with P1 and P1 replicates to P2 (see the
>>>t_replicate() function in TM module). But again, this makes sense only
>>>if using cache for usrloc
>>
>> Indeed.  I see now.  Thanks.
>>
>>
>> - Jeff
>>
>>
>>>
>>>
>>>Regards,
>>>Bogdan
>>>
>>>--
>>>Bogdan-Andrei Iancu
>>>OpenSIPS Event - expo, conf, social, bootcamp
>>>2 - 4 February 2011, ITExpo, Miami,  USA
>>>OpenSIPS solutions and "know-how"
>>>
>>>
>>>_______________________________________________
>>>Users mailing list
>>>Users at lists.opensips.org
>>>http://lists.opensips.org/cgi-bin/mailman/listinfo/users
>>
>> _______________________________________________
>> Users mailing list
>> Users at lists.opensips.org
>> http://lists.opensips.org/cgi-bin/mailman/listinfo/users
>>
>
>_______________________________________________
>Users mailing list
>Users at lists.opensips.org
>http://lists.opensips.org/cgi-bin/mailman/listinfo/users