[SR-Users] NatHelper's Keep-alive timeout behind proxy

[Asgaroth]

On 17/05/18 14:23, Daniel Tryba wrote:
>
> I noticed the same problem (most important with multiple registrars I
> didn't like paritioning logic) and my solution was to send my own OPTIONS.

We have it running with multiple (3) registrars that replicate with dmq 
usrloc, works perfectly so far for us.

> In your favorite scripting language, select all location entries and
> construct an OPTIONS and send that directly to the correct proxy with a
> preloaded Route (based on Path headers inserted on the proxy). In the
> response I defined a valid reply as a response code other than 408 and
> below 500.
>

I'm not sure I'm following your logic here. The registrars with 
nathelper loaded are already doing what you describe above. They are 
actually sending the options keepalive using path from the 
location/registrar module. The options messages are getting to the ua's 
and that part is working as intended.

The part we're having an issue with is when we enable the 
keepalive_timeout module parameter. My understanding is that with this 
module parameter enabled, if a ua is sent a keepalive message and 
nathelper does *not* recieve a response back from the client after 
predefined number of attempts, then nathelper will flush the contact 
from location as it deems it as "down".

When the options message is sent via the correct proxy to a ua that is 
"down", the proxy retransmits the options 3 - 4 times to the ua, and 
because the ua does not respond (its down), kamailio generates a 408 
Request Timeout response back to the registrar for that keepalive 
request. The behaviour of the proxy is expected behaviour under normal 
conditions, however, it causes an issue with this sceanrio in that the 
408 generated by the proxy and sent back to the registrar is interpreted 
by the nathelper module as a successfull response from the ua and 
therefor does not remove the contact from the location table, even 
though the request had timed out.

Are you saying that you manualy remove your contacts from the database 
using your custom script based on the responses back? We are currently 
running in memory only mode, so it makes it a little more difficult.

I think what would be nice is if nathelper had a similar option to what 
dispatcher has, in the sense that you could defined, via module 
parameter, what response code would be deemed as a "failure" in 
contacting the ua, that way, in this scenario, we could say that a 408 
timeout should be considered as identicle to a non-response from the ua.

I cant think of how to disable sending these 408 from the proxy for 
these specific types of messages, or, if this is even possible.