Can you publish your measurements somewhere? Adrian
On Apr 5, 2012, at 10:54 PM, Jock McKechnie wrote: > Thanks Vlad; > > I realised that we're several releases behind, so I compiled 1.7.2 > with memory debugging built in. Today our traffic was relatively low > so I was mostly averaging ~800 calls concurrently. So far so good. > It's possible that the update to 1.7.2 has done the trick. > > I have noticed that when more heavily loaded we start seeing a number > of errors coming out of the dialog module, like: > log_next_state_dlg: bogus event 7 in state 2 for dlg .... > With the event always remaining 7, but the state moving about between > 1 and 2. Is the dialog module not able to keep up? > > - JP > > On Thu, Apr 5, 2012 at 3:58 AM, Vlad Paiu <[email protected]> wrote: >> Hi, >> >> It might be a memory leak somewhere. >> >> Can you please follow the following tutorial in order to get a memory dump, >> so we can further investigate this ? >> http://www.opensips.org/Resources/DocsTsMem >> >> Regards, >> >> Vlad Paiu >> OpenSIPS Developer >> http://www.opensips-solutions.com >> >> >> >> On 04/05/2012 01:00 AM, Jock McKechnie wrote: >>> >>> In addition to my MediaProxy loading learnings I'm also discovering >>> that there appears to be some fiddling to get OpenSIPS to handle this >>> kind of loading as well. I have a fairly simple configuration which >>> engages media proxy and forwards packets onto a single destination >>> proxy (who then sends the calls onto the carrier). >>> >>> When I load up OpenSIPs with around 1400 inuse_transactions it'll >>> click along for an hour or two and then it'll bomb out with: >>> WARNING:core:fm_malloc: Not enough free memory, will atempt defragmenation >>> ERROR:core:do_action: memory allocation failure >>> <OpenSIPS threads shutting down here> >>> ERROR:nat_traversal:save_keepalive_state: failed to open keepalive >>> state file for writing: Permission denied >>> CRITICAL:core:sig_alarm_abort: BUG - shutdown timeout triggered, dying... >>> >>> I'm using the Debianised 1.6.4-notls under a VM with 512MB of RAM. I >>> have OpenSIPs set to allow 128MB in /etc/default/opensips. We have >>> almost two hundred OpenSIPS proxies all running on 64M and I've >>> _never_ managed to make one croak, but I imagine this one is much more >>> heavily loaded due to the volume of calls along with the use of the >>> dialog module for mediaproxy. >>> >>> From when the proxy is first fired up, it's handling a fairly >>> consistent 1000-1400 transactions, but it takes at least an hour for >>> it to go over, which suggests to me that memory isn't being released >>> properly. I'd be happy to throw more memory at the issue... but if >>> it's not releasing then I'm left to wonder if there's a flaw that is >>> always going to cause this problem. >>> >>> Am I barking up the wrong tree here, or is this known and there's a >>> reasonable solution, or? >>> >>> My config, for reference, is here: http://pastebin.com/ZiWbK3GJ >>> >>> As always, thank you very much; >>> >>> - Jock >>> >>> _______________________________________________ >>> Users mailing list >>> [email protected] >>> http://lists.opensips.org/cgi-bin/mailman/listinfo/users >> >> >> _______________________________________________ >> Users mailing list >> [email protected] >> http://lists.opensips.org/cgi-bin/mailman/listinfo/users > > _______________________________________________ > Users mailing list > [email protected] > http://lists.opensips.org/cgi-bin/mailman/listinfo/users > _______________________________________________ Users mailing list [email protected] http://lists.opensips.org/cgi-bin/mailman/listinfo/users
