On 11/18/2011 03:16 PM, Andres wrote:
My logic tells me:
- If its not hitting your xlog in your register section then its not
registering to this server
The registrar is running inside an OpenVZ container, so the first possibility I investigated was that someone had inadvertently cloned the container on the same hardware node and that there was an IP address conflict. However, this turned out not to be the case. I also double-checked the MAC addresses on the registrations within the successful interval vs. outside of it.
- If your packet capture is showing a successful registration and
you cannot see it registered in this server then its registering somewhere else.
That stands to reason, but I cannot find any evidence apart to support that viewpoint, nor anything about the packet capture (e.g. different hardware or instance signatures) that would seem to corroborate that.
- Is it possible that your server is simply re-routing the packet
to another server before it hits your xlog statement and thus the client is registering there?
I entertained that possibility, but can't find any indication of how this would be possible, particularly in an arbitrary, small number of cases for very brief periods.
A full packet capture in and out of the server might give you a hint, but if you have thousands of customers that might be tough.
I took a packet capture at the edge router of the network as well as on the hardware node itself, in addition to the container. All of them tell the same story.
On the other hand, if you have thousands of customers, I would venture to guess that you have more than one server where users can register and that might explain why the trace shows a successful registration.
There is just one registrar -- the one under discussion. It's the main reason Kamailio is used here; it is the only OSS registrar I know of that has enough throughput capacity to sustain thousands of devices banging on it with relatively short re-registration intervals.