Then rework the logs. Do you have more than one log per registration? If yes, use only one. If you already have a single log per registration, then try reducing the size. Experiment to see where the system breaks. Try new hardware. Decrease the number of workers to see if it has an impact.
-ovidiu
On Thu, Sep 15, 2022 at 11:30 Amit amit@brytecall.com wrote:
Hi Henning,
No we are using a local graylog server, not cloud.
I suppose if the bottleneck is log generation, it doesn’t matter how or where they are handled/sent.
Amit
*From: *Henning Westerholt hw@gilawa.com *Date: *Thursday, September 15, 2022 at 9:53 AM *To: *Amit amit@brytecall.com *Cc: *Kamailio (SER) - Users Mailing List sr-users@lists.kamailio.org *Subject: *RE: [SR-Users] Log levels severely impacts performance
Hello Amit,
this sounds strange. If you send them remotely it should not really affect the operation anymore.
Maybe it’s a special limitation of the cloud server, if you use a public cloud service provider, but just guessing here.
Cheers,
Henning
--
Henning Westerholt – https://skalatan.de/blog/
Kamailio services – https://gilawa.com
*From:* Amit amit@brytecall.com *Sent:* Thursday, September 15, 2022 3:16 PM *To:* Henning Westerholt hw@gilawa.com *Cc:* Kamailio (SER) - Users Mailing List sr-users@lists.kamailio.org *Subject:* Re: [SR-Users] Log levels severely impacts performance
Hi Henning,
We are actually sending the logs directly to our local graylog server via the $programname property in rsyslog, and we’ve disabled writing Kamailio logs to disk (i.e. only in memory by setting Storage=volatile in rsyslog.conf) but that didn’t seem to help with the application socket receive buffer backlog.
Keep in mind that for testing purposes, I am blasting Kamailio with a fixed 2000 simultaneous registrations, so at log level 2 it really struggles and the socket buffer quickly fills up. Reducing the log level to 1 allows Kamailio to rip through the requests and the buffer stays at 0.
Amit Nir, MBA
Office: 305.999.0911
www.BryteCall.com
*From: *Henning Westerholt hw@gilawa.com *Date: *Thursday, September 15, 2022 at 9:01 AM *To: *Amit amit@brytecall.com *Cc: *Kamailio (SER) - Users Mailing List sr-users@lists.kamailio.org *Subject: *RE: [SR-Users] Log levels severely impacts performance
Hello,
how much are you actually logging? As already suggested, maybe just need to decrease the amount of log messages.
If you need this much logging, you can of course forward it to a dedicated log server over the network.
Cheers,
Henning
--
Henning Westerholt – https://skalatan.de/blog/
Kamailio services – https://gilawa.com
*From:* Amit amit@brytecall.com *Sent:* Thursday, September 15, 2022 1:33 PM *To:* Henning Westerholt hw@gilawa.com *Cc:* Kamailio (SER) - Users Mailing List sr-users@lists.kamailio.org *Subject:* Re: [SR-Users] Log levels severely impacts performance
Hi Henning,
Thanks for the suggestion.
Disk I/O was our initial assumption on another Kamilio instance we are running.
To test this theory we installed Kamailio on separate hardware and configurations with faster underlying SSD disks, however we are seeing the same crippled performance and congestion in the applications socket receive buffer when LOG level is set to INFO (2).
Amit Nir, MBA
Office: 305.999.0911
www.BryteCall.com
On Sep 15, 2022, at 6:14 AM, Henning Westerholt hw@gilawa.com wrote:
Hello,
additionally, try to see if you have some general I/O issues, e.g., by observing io-wait CPU, looking to I/O transaction times with monitoring tools etc..
Maybe there are other services also causing excessive I/O.
Cheers,
Henning
--
Henning Westerholt – https://skalatan.de/blog/
Kamailio services – https://gilawa.com
*From:* sr-users sr-users-bounces@lists.kamailio.org *On Behalf Of *Daniel-Constantin Mierla *Sent:* Wednesday, September 14, 2022 2:25 PM *To:* Kamailio (SER) - Users Mailing List sr-users@lists.kamailio.org; Amit amit@brytecall.com *Subject:* Re: [SR-Users] Log levels severely impacts performance
Hello,
excessive logging is known to impact the performances. If you have many log messages printed from config, try to reduce them.
Performance could be also a matter of how many child processes you created, via the children global parameter.
Cheers, Daniel
On 14.09.22 04:18, Amit wrote:
Hello,
We’ve been running Kamailio 5.5.0 for a while now and recently ran into an issue where the application buffer would get full and Kamailio has a lot of difficulty keeping up with REG requests. This severely impacted Kamailio’s ability to perform and resulted in many REG timeouts.
Running ‘watch netstat -ulpn’ revealed the buffer issue.
We usually run the production system at log level INFO (2) and had not ran into any issues in the past.
When the buffer issue occurred and after some testing, we discovered that changing the log level to NOTICE (1) was sufficient to bring the system back into operational status.
For example, at log level INFO (2) we seem to achieve ~150-200 REG/s while at log level NOTICE (1) we reach ~500-700 REG/s.
I did find this article in the wiki and tried using the “-“ in front of the log path but this didn’t seem to help.
http://www.kamailio.org/wiki/tutorials/3.2.x/syslog
Even writing the logs directly to our external log server and preventing them from being written to the local disk doesn’t seem to alleviate the issue when the log level is set to INFO (2).
Does anyone have any experience with this or have any suggestions? We are finding it difficult to scale as a result of this issue.
Thank you in advance.
Amit
Kamailio - Users Mailing List - Non Commercial Discussions
- sr-users@lists.kamailio.org
Important: keep the mailing list in the recipients, do not reply only to the sender!
Edit mailing list options or unsubscribe:
--
Daniel-Constantin Mierla -- www.asipto.com
www.twitter.com/miconda -- www.linkedin.com/in/miconda
Kamailio - Users Mailing List - Non Commercial Discussions
- sr-users@lists.kamailio.org
Important: keep the mailing list in the recipients, do not reply only to the sender! Edit mailing list options or unsubscribe: