Hello Andrew,

indeed, the difference is only 1.4k or so memory. This is probably just an effect of the different time when you did the dump.

Yes, an Github issue would be great. Add this dump as well https://paste.ubuntu.com/p/85KYg7gD2z/

Would be perfect if you could reproduce it with current git master, though (if the app_ruby problem is also solved).

Cheers,

Henning

Am 06.08.19 um 09:17 schrieb Andrew White:
Very interesting results!

11pm last night:
https://paste.ubuntu.com/p/qMz2tN7bcR/

2pm today:
https://paste.ubuntu.com/p/QDmTwpV83X/

The growth is pretty small. The only variance is disabling the topoh module.

I’m not big on C, but I think this indicates a leak in the topoh module, does this sound right?

I’m happy to put in an issue on GitHub if that’s necessary here.

Cheers,

Andrew

On 5 Aug 2019, at 11:30 pm, Andrew White <andrew@uconnected.com.au> wrote:

Hey Henning,

Unfortunately I’ve had some troubles with app_ruby in dev-7. Interestingly, dev-7 shows as “dirty” too:

version: kamailio 5.3.0-dev7 (x86_64/linux) bd477f-dirty

Here’s the ansible playbook I wrote to pull the specific version, there’s no fancy patching or anything going on:

- name: Download kamailio
git:
version: "{{ kamailio_version }}"
dest: /usr/src/kamailio
- name: Roll back to specific commit
command: "git reset --hard {{ kamailio_commit }}"
args:
chdir: /usr/src/kamailio
when: kamailio_commit is defined
- name: Build kamailio from source
command: "{{ item }} chdir=/usr/src/kamailio"
with_items:
- make include_modules="app_ruby db_redis" cfg
- make all
- make install
args:
creates: /usr/local/sbin/kamailio

Either way, I’ve removed the topoh module from my production (dev-4) machine, and re-taken memory information. I’ll give it about 12 hours and see where memory is sitting. I figure this should isolate if it’s a topoh issue or not.

Thanks!

Andrew


On 5 Aug 2019, at 10:46 am, Andrew White <andrew@uconnected.com.au> wrote:

Hi Henning,

That is interesting, I don’t know what the dirty version is. My deploy script I use for Kamailio pulls 6d43eacd6d5b78fe857267e70b7fa85519a7d5b6 from GitHub, as this had a fix in place Daniel did for app_ruby a while back. I don’t recall ever putting on any custom patches or modules. I compile from source on build from that git commit using:

make include_modules="app_ruby db_redis" cfg

I’ll deploy dev7 tonight and report back!

Thanks,

Andrew

On 4 Aug 2019, at 8:28 pm, Henning Westerholt <hw@skalatan.de> wrote:

Hi Andrew,

dev4 version - I see. I quickly looked into the topos module commits, but there have been not that much changes there. But can you try nevertheless to update to the latest git master (e.g. dev7) version on your test setup?

What about the "dirty" version flag, do you have some own patches or modules that you use?

Cheers,

Henning

Am 04.08.19 um 01:42 schrieb Andrew White:
Hey Henning,

It’s currently version: kamailio 5.3.0-dev4 (x86_64/linux) 6d43ea-dirty. 

I don’t think there’s any extra messages. The actual dispatching is simple, here’s the code:

  1. if destination then
  2. KSR::PV.sets("$du", destination['sip_uri'])
  3. pbx = destination['display_name']
  4. KSR.info("Chose #{pbx} as inbound PBX, as it's received #{destination['count'].to_i}/#{destination['total_count'].to_i} (#{destination['calculated_percentage'].to_i}%) of calls in this ruleset, where its max is #{destination['percentage'].to_i}%")
  5. end

The code before that just pulls data from redis and returns a ruby hash with where we should send the call, so no kamailio variables set at all.

Thanks!

Andrew

On 3 Aug 2019, at 8:03 pm, Henning Westerholt <hw@skalatan.de> wrote:

Hi Andrew,

interesting insight from the logs indeed. According to your changes this looks indeed like related to the topoh module or custom dispatcher code. Which version of Kamailio do you use?

Are you sending out some messages /replies as well from your custom dispatcher app_ruby logic?

Cheers,

Henning

Am 03.08.19 um 09:32 schrieb Andrew White:
Hi all,

I’ve done two outputs about 8 hours/3200 calls in between:

First one (about 5 hours since last restart):


Second one (about 8 hours after the first, 3200 odd cals later):


It looks like roughly 7MB of extra memory is being used on that process, which is 1/4 active workers (I can see all of them have grown roughly the same amount in that time). Looking at the memory status, there appear to be about 14,000 more lines. A cursory glance shows about 6k of those lines with msg_translator.c (compared to 1k in the first).

I wonder if maybe this could be related to my use of the topoh module, given every message is now being touched by it to mask the origin?

Thanks!

Andrew

On 3 Aug 2019, at 10:07 am, Andrew White <andrew@uconnected.com.au> wrote:

Hi Daniel/Henning,

Thanks so much for your quick responses, they’re very much appreciated!

Daniel: I don’t believe so. I’ve just checked over the code, and I only have 9 instances of PV.sets throughout the entire script. Additionally, this issue appears to have only cropped up in the last week or two (I didn’t monitor PKG memory at the time, but we didn’t have any memory related crashes before that point), which leads me to believe it’s related to my config code, the adding of the topoh module or the removal of the dispatcher module. 

Within that config code, there’s only one additional Kamailio variable set, which is part of my replacement/custom dispatcher (app_ruby):

KSR::PV.sets("$du", destination['sip_uri’])
For now, I’ve taken the output of a mem dump of one of the production workers, ps output, etc. I’ll leave that few a few hours, monitor to ensure there is a measurable memory increase, and send through the logs if that’s ok?

Thanks!

Andrew

On 2 Aug 2019, at 11:44 pm, Daniel-Constantin Mierla <miconda@gmail.com> wrote:

Hello,

are you defining a lot of kamailio variables in your ruby script? In other words, are you using variables with different name, like $var(xyz) or $sht(a=>xyz), where xyz is passed from/computed in ruby script and is changing depending of the sip message?

Cheers,
Daniel

On 01.08.19 15:34, Andrew White wrote:
Thanks Daniel, you’re fantastic!

I have 4 children/workers configured with -m 128 -M 32. The machine in question has 512MB of memory, 1 core and 1GB swap on an SSD.

I restarted Kamailio with memlog=1 and I’ve been sending batches of 30 calls in. I’ve noticed 4 of the 13 Kamailio processes going up in memory after each batch, which I suspect to be the primary children/workers. Immediately post restart:

root     28531  0.7  5.5 329368 27196 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28532  0.6  4.9 329368 24528 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28533  0.6  5.5 329368 27244 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28534  0.7  5.4 329368 26788 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32

After about 90 calls:

root     28531  0.0  6.7 330688 32948 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28532  0.0  6.5 330560 32264 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28533  0.0  6.5 330556 32272 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28534  0.0  6.6 330564 32592 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32

None of the other 9 Kamailio processes are increasing at all.

I ran corex.pkg_summary against one of them and got the following dump:


I can see a lot of allocation to pvapi.c, does this indicate maybe I’m setting PVs that need to be unset?

Here’s another after another 60 calls:


root     28531  0.0  6.9 330820 33928 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28532  0.0  6.7 330692 33352 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28533  0.0  6.7 330688 33280 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32
root     28534  0.0  6.7 330696 33192 ?        Sl   22:48   0:00 /usr/local/sbin/kamailio -DD -P /var/run/kamailio/kamailio.pid -f /etc/kamailio/kamailio.cfg -m 128 -M 32

The only changes I’ve made on this config over the last couple of weeks (since I saw this issue) is removing the dispatcher module and adding in a small function in app_ruby (which I already use) to query redis (which I also already use from app_ruby and make a heap of queries per call) for some values and write $du manually. I also added in the topoh module.

It also makes a lot of sense to me to monitor the individual processes rather than the aggregate. Is there a way to identify simply from bash what processes are workers programmatically? I’d like to monitor just those individually in my monitoring.

Thanks!

Andrew


On 1 Aug 2019, at 8:24 pm, Daniel-Constantin Mierla <miconda@gmail.com> wrote:

Hello,

if it is pkg, then you have to see which process is increasing the use of memory, because it is private memory, specific for each process. The sum is an indicator, but the debugging has to be done for a specific process/pid.

Once you indentify a process that is leaking pkg, execute the rpc command:

  - https://www.kamailio.org/docs/modules/devel/modules/corex.html#corex.rpc.pkg_summary

When that process is doing some runtime work (e.g., handling of a sip message), the syslog will get a summary with used pkg chunks. Send those log messages here for analysis. You have to set memlog core parameter to a value smaller than debug.

Cheers,
Daniel


On 01.08.19 03:43, Andrew White wrote:
Hi all,

I had a Kamailio crash the other day, and some debugging showed I ran out of PKG memory.

Since then I’ve run a simple bash script to compile the amount of memory used by all child processes, effective /usr/local/sbin/kamcmd pkg.stats | grep real_used summed together. I’ve graphed out the data, and there’s a clear growth of PKG memory going on, mostly increasing during our busier daytime hours.


Based on this, I suspect either a module loaded or something within my app_ruby conf is leaking memory.

I’ve been reading through https://www.kamailio.org/wiki/tutorials/troubleshooting/memory, but I’m a bit nervous, as I’m not really a C/deep memory type of guy. I can see a GDB script I can attach to Kamailio, but is that going to use significant resources to run or impact the running process? Is there a newer/better/alternative way to do this, and to help me break this down?

Thanks!

Andrew

_______________________________________________
Kamailio (SER) - Users Mailing List
sr-users@lists.kamailio.org
https://lists.kamailio.org/cgi-bin/mailman/listinfo/sr-users
-- 
Daniel-Constantin Mierla -- www.asipto.com
www.twitter.com/miconda -- www.linkedin.com/in/miconda

-- 
Daniel-Constantin Mierla -- www.asipto.com
www.twitter.com/miconda -- www.linkedin.com/in/miconda



_______________________________________________
Kamailio (SER) - Users Mailing List
sr-users@lists.kamailio.org
https://lists.kamailio.org/cgi-bin/mailman/listinfo/sr-users
-- 
Henning Westerholt - https://skalatan.de/blog/
Kamailio services - https://skalatan.de/services

-- 
Henning Westerholt - https://skalatan.de/blog/
Kamailio services - https://skalatan.de/services



-- 
Henning Westerholt - https://skalatan.de/blog/
Kamailio services - https://skalatan.de/services