11 Oct
2019
11 Oct
'19
6:02 a.m.
Branch: refs/heads/1.6.0
Home: https://github.com/tribe29/checkmk
Commit: d1146913f77c3674a675bd1f0bd59a24e1b24acf
https://github.com/tribe29/checkmk/commit/d1146913f77c3674a675bd1f0bd59a24e…
Author: Ronny Bruska <ronny.bruska(a)tribe29.com>
Date: 2019-10-11 (Fri, 11 Oct 2019)
Changed paths:
A .werks/10070
M agents/cfg_examples/systemd/check_mk@.service
Log Message:
-----------
10070 FIX systemd: Fixed possible memory leak related to KillMode in unit file
We have seen cases in which the parameter "KillMode=process" of the unit file
"check_mk@.service" causes a memory leak. This was first seen on Debian Buster
but could occur in the future on other operating systems too.
The cached check runs that are forked in the backround seem to cause systemd
not to discard the service once it's done. If you run:
C+:
RP:systemd-cgls -au system-check_mk.slice
C-:
you can see a lot of old services and task_struct objects in slab growing
constantly. You can see those processes also on eg. Debian Stretch but there
they should have no effect on task_struct objects.
To fix this, the parameter "KillMode=process" was changed to
"Type=forking".
To change the unit file:
C+:
RP:systemctl stop check_mk.socket
RP:systemctl disable check_mk.socket
C-:
Copy the new check_mk@.service unit file to /etc/systemd/system/.
You can download the file within "Monitoring Agents" - "Agent files"
or find it in site user context ~/share/check_mk/agents/cfg_examples/systemd/.
C+:
RP:systemctl daemon-reload
RP:systemctl enable check_mk.socket
RP:systemctl start check_mk.socket
C-:
Change-Id: Ic5ed0916c323b22eb7339762d32c330639dc1869