Hello Henning, Due to some analysis (thanks for the activity to Enrico Bandiera and Giacomo Vacca) we found a bug in dmq module , internally made a quick and dirty patch and opened a issue on GitHub !
Cheers
Il giorno sab 26 gen 2019 alle 19:19 * Paolo Visintin - evosip.cloud paolo.visintin@evosip.cloud ha scritto:
Hello Henning, Correctly understood, this is exactly what I've experienced (kamailio 5.2.0) I'll absolutely make a new lab-test and strace !
Thanks for your suggestion! Best regards
Il giorno sab 26 gen 2019 alle ore 19:05 Henning Westerholt < hw@kamailio.org> ha scritto:
Am Samstag, 26. Januar 2019, 19:00:32 CET schrieb * Paolo Visintin -
I'm using DMQ in order to share :
- htable
- usrloc
for usrloc seems everything working as expected.
on htable I have noticed, after some stresstest made with sipp (25 cps / 600 concurrent calls) that one of the nodes (or sometimes both) eat a
lot
of cpu (300%) after stresstest ended [all processes idle except f4 dmq handlers]; I tried also to change some parameters on dmq module like worker_usleep with no changes at all.
Hello Paolo,
just commenting on the first issue:
So I understood you correctly, after the stress-test you observe a not ending CPU load on the Kamailio server, even without load?
Maybe you can have a look on the CPU consuming processes by attaching e.g. "strace" to them at the next occasion, to see what they are doing.
Best regards,
Henning
At last, using kamailio in a kubernetes system I have several kamailio instances that pops-up and then tiers down with different ips; for this reason dmq seems not able to delete peers, adding, for every defunct
node a
warning like this : router-3 17(34) ERROR: dmq [notification_peer.c:599]: notification_resp_callback_f(): deleting server sip:172.28.1.211:5062;status=active because of failed request router-1 router-sr 17(33) ERROR: dmq [notification_peer.c:599]: notification_resp_callback_f(): deleting server sip:172.28.1.213:5062;status=active because of failed request
Il giorno sab 26 gen 2019 alle ore 01:43 Alex Balashov <
abalashov@evaristesys.com> ha scritto:
I think the realistic answer is that it's getting to be that way. I think DMQ is now the recommended way to share usrloc and htable across multiple hosts. DMQ+dialog is still a work in progress. Not sure about DMQ+some other new stuff.
On Fri, Jan 25, 2019 at 04:21:43PM +0000, Canuck . wrote:
Before DMQ there were other ways to do similar things. Is DMQ now considered the best way? Assuming I am using recent stable?
Are there any good guides on clustering Kamailio? I am not finding
a
lot of good info on that subject. Specifically, for multi
datacenter
active active where things like floating IP's and keepalived are not really an option.
-- Henning Westerholt - https://skalatan.de/blog/ Kamailio services - https://skalatan.de/services Kamailio security assessment - https://skalatan.de/de/assessment
--
paolo visintin direttore tecnico ------------------------------ timenet srl | www.timenet.it | tel 05711738000 assistenza tecnica: assistenza@timenet.it ufficio commerciale: sales@timenet.it ufficio amministrativo: contabilita@timenet.it twitter.timenet.it | linkedin.timenet.it http://timenet.it/email/redirect.php