Linux Container on proxmox

Agree, I have same problem,
I'am use proxmox with LXC containers.

I created custom UserParameter for solve it:
https://medium.com/@kvapss/zabbix-solve-memory-monitoring-issue-inside-lxc-containers-98ddf191051c

I have the same problem on LXC machines. Zabbix agent returns total host memory instead of total container memory.

While it's possible to create custom UserParameter, we can't override default keys with UserParameter, so we have to maintain different templates for container. It'd be better to be able to use a generic one and get the correct value

Hi, I've found out that CPU Load collecting wrong too, so I've added new UserParameter into my zabbix template:
https://github.com/kvaps/zabbix-linux-container-template
Please use it

I here thought in some cases this behavior can play into my hands, for example when I want to monitor physical host but want to run zabbix-agent inside the container.
In my opinnion this is good idea to have option inside `zabbix-agentd.conf` for set what data exactly needed, is that data from the container environment or from the physical host.

Any comments from Zabbix Team?

Zabbix Team, please, look at this issue. Repair is probably quite easy  Thanks.

Zabbix agent is monitoring memory over sysinfo() syscall. Looks like Linux namespaces are still not wrapping what is should be returned in the container. Even if zabbix will switch from sysinfo() syscall to what provides /proc/memoryinfo still it will be the issue on kernel side. I don't know how it is with latest kernel versions but at least Ubuntu 4.4.0-141-generic has this issue still not solved.
IMO this problem needs to be sorted out on kernel side because sysinfo() syscall shpould report some values correctly.

Same problem too

If someone in Zabbix is going to spend some time on that issue IMO better would be invest that time into fixing what returns sysinfo() syscall and submit that patch to the kernel source tree.

Nothing inside lxc zone like on Solaris shout present any data from global zone.

I researched similar behavior in case if Docker is used and made some conclusions.

Inside Docker containers /proc/meminfo and sysinfo() are returning values from host, not from container. top and free also return values from host - as they fetch values from /proc/meminfo (and other files like cpuinfo, diskstats, stat, swaps, uptime). Docker does not provide any substitutes for these files like LXCFS does.
Container's metrics (that respect its limits) can be fetched from cgroups metrics in /sys/fs/cgroup/memory/ (memory.stat and other memory.* files). This path is valid for cgroups v1.
In the nearest future cgroups v2 will become mainstream (its already used in Podman) and cgroups sysfs file structure and hierarchy will be different.
Users can monitor individual container's metrics through host, by using their containerization platform's APIs and tools.
From the other side: agent2 with https://www.zabbix.com/integrations/docker exists.

It will be undesirable to start fetching values from /proc/meminfo instead of sysinfo - because it will introduce discrepancy and confusion:

• LXCFS is a feature provided by LXC. Docker/Podman does not provide any substitutes for /proc/ files. Therefore if agent will fetch values only from /proc/meminfo, the discrepancy between behavior of different containers will be introduced.
• User may want to containerize agent and collect metrics from host (maybe even without using LXCFS). User will not have choice if sysinfo values will be simply replaced with /proc/meminfo values.

My conclusion:

• This is not a bug in Agent, it is just a small nuisance of containerization.
• New agent metrics can be implemented to fetch container's metrics using cgroups' sysfs. However, this is new feature implementation and it is out of this issue's scope.

@dgoloscapov As I wrote IMO this is not per se zabbix bug and it should be solved outside of the zabbix. I'm not sure who is maintaining namespaces code in Linux kernel code but this issue should be fixed there and only there. If it will be fixed in kernel area all typed of containerisations will be possible to monitor OOTB. I would even abandon provide metrics based on sysfs content because when that issue will be solved on kernel layer that code will be only legacy.

Tengo el mismo problema. Espero que tenga pronto alguna solución oficial.

saludos

According to github issue, lxc cant/wont change sysinfo response, so "the bug should be solved outside" is not possible. We can call this situation "development deadlock"