On 02/04/2014 12:54 PM, Olle E. Johansson wrote:
And you assume a lot of things - that your invites are sent to the same server, which may not be the case. DNS lookups and load balancing/failover is done on a per transaction basis.
In fairness, RFC 3578 does offer a fair bit of commentary on this caveat:
However, having subsequent INVITEs routed in different ways brings some problems as well. The first INVITE, for instance, might be routed to a particular gateway, and a subsequent INVITE, to another. The result is that both gateways generate an IAM. Since one of the IAMs (or both) has an incomplete number, it would fail, having already consumed PSTN resources.
[...]
Routing in SIP can be controlled by the administrator of the network. Therefore, a gateway can be configured to generate SIP overlap signalling in the way described below only if the SIP routing infrastructure ensures that INVITEs will only reach one gateway. When the routing infrastructure is not under the control of the administrator of the gateway, the procedures of Section 2 have to be used instead.
And, while I agree that this is ridiculous and is in conflict with the basic spirit of SIP, somehow 3578 did become an RFC... I am as puzzled by that as you may be. :-)
-- Alex