[Users] uac_redirect out of memory?

[Greg Fausak]

Daniel,

I figured the memory setup was just insufficient.
I have set up a redirect server, I don't have subscribers.  My
routing script is simple, looks something like:

route
{
	checkfor misc errors;
	looseroute;

if (src_ip == 10.88.128.115)
{
   if (is_method("INVITE"))
   {
    t_on_failure("3");
    record_route();
    rewritehostport("10.88.129.253:5060");
    if (!t_relay()) {
     sl_reply_error();
    }
    return;
   }
}
else if(src_ip == 10.88.129.253)
{
   if (is_method("INVITE") || is_method("CANCEL") || is_method("BYE"))
   {
    record_route();
    rewritehostport("10.88.128.115:5060");
    if (!t_relay()) {
     sl_reply_error();
    }
    return;
   }
   sl_send_reply("405", "method not supported"); return;
}
  sl_send_reply("403", "I am not familiar with your IP address  
Maynard"); return;
}

failure_route[3]
{

if (t_check_status("3[0-9][0-9]"))
{
   get_redirects("*");
   serialize_branches(0);
}
if (next_branches())
{
   t_on_failure("3");
   t_relay();
   return;
}
t_reply("404", "No branches left");
}

That's it.  When I apply a load I start getting memory shortages.
I have been able to get around the problem using -m.  I was looking
for a better way other than guessing.  If I knew how many bytes are
consumed per pending transaction, or, if I could tell how much shared  
memory
is being used at any given moment, I could do a better job at providing
the correct -m value.

---greg


On Sep 4, 2006, at 5:11 AM, Daniel-Constantin Mierla wrote:

> Hello,
>
> if you get "out of memory" at peak and then it recovers, it is not  
> a memory leak, just insufficient shared memory. Use -m command line  
> option to tune the value. By default it is 32MB.
>
> The size of memory you have to allocate depends on how many  
> subscribers online you have. The user location cache uses shared  
> memory as well.
>
> Cheers,
> Daniel
>
>
> On 09/02/06 00:54, Greg Fausak wrote:
>> We've been doing a little bit
>> of stress testing on the auc_redirect module.
>> Throwing 60 calls per second at it.  After quite
>> a while I start getting error messages like:
>>
>> Sep  1 15:33:43 www openser[74528]: ERROR: sip_msg_cloner: cannot
>> allocate memory
>> Sep  1 15:33:43 www openser[74499]: redirect.g2.sbsqa2.com-LOOSE:
>> time_t=1157142823 ^LINE=47 ^FILE=/var/local/ser/ROUTLET/config.pre
>> ^call_id=5723-28564 at 10.88.129.253 ^cseq=3
>> ^contact=<sip:18001002132 at 10.88.129.253>
>> ^from=sip:+18001002132 at 10.88.129.253 ^fromtag=18001002132
>> ^to=sip:+16661012126 at 10.88.128.116 ^totag=5724 ^method=BYE
>> ^ruri=sip:+116661
>> 012126 at 10.88.129.251:5060 ^messageid=263458^remark=Doing it
>> Sep  1 15:33:43 www openser[74528]: ERROR: new_t: out of mem:
>> Sep  1 15:33:43 www openser[74592]: redirect.g2.sbsqa2.com-LOOSE:
>> time_t=1157142823 ^LINE=47 ^FILE=/var/local/ser/ROUTLET/config.pre
>> ^call_id=5724-28564 at 10.88.129.253 ^cseq=3
>> ^contact=<sip:18001002133 at 10.88.129.253>
>> ^from=sip:+18001002133 at 10.88.129.253 ^fromtag=18001002133
>> ^to=sip:+16661012127 at 10.88.128.116 ^totag=5725 ^method=BYE
>> ^ruri=sip:+116661
>> 012127 at 10.88.129.251:5060 ^messageid=263192^remark=Doing it
>> Sep  1 15:33:43 www openser[74528]: ERROR: t_newtran: new_t failed
>> Sep  1 15:33:43 www openser[74499]: ERROR: sip_msg_cloner: cannot
>> allocate memory
>> Sep  1 15:33:43 www openser[74592]: ERROR: new_t: out of mem:
>> Sep  1 15:33:43 www openser[74499]: ERROR: new_t: out of mem:
>> Sep  1 15:33:43 www openser[74592]: ERROR: t_newtran: new_t failed
>>
>> in the log file.  I have seen similar problems with regular t_relay
>> when we approach 200
>> calls per second.  Is this a shared memory problem?  How much  
>> shared memory to
>> I need to set up?
>>
>> Have a great weekend!
>>
>> ---greg
>>
>>