LXC containers shows host's load average

[EmilioMoreno]

Hi there.
I remember reading in another thread in this forum that. this was not an Proxmox issue but some kind of bugs between lxcfs and cgroups2 or something like that but I am now unable to find the thread. For me, as all my containers are monitored from custom agents inside the container for load and memory this simply blocks my upgrade path, and I will be in pve6 until this gets addressed (if gets addressed).

The alternative for me is develop/deploy a new monitoring system running on the physical Proxmox hosts for more than 200 containers, which is currently not an option.

@Dunuin if you are using zabbix agent inside the container (I am also for some of my nodes), I have just created my own items for reading memory. The right memory values inside the container can be found in /proc/meminfo so you just create an item which reads this, using
key = vfs.file.contents[/proc/meminfo]
and then just create dependent items which parse the values of the original item to generate the free/used/avalable/total items.

 

[Dunuin]

Thanks, will have a look at that when I find some time.

 

[yswery]

Did anyone end up finding a fix for this issue where the LXC containers show the hostnode CPU use

 

[Dunuin]

No, not really.

 

[EmilioMoreno]

anyone in 7.3 who can confirm if this is address on that version?

 

[tomee]

Hasn't changed for me. I still see host cpu usage at the idle container.

 

[tuxillo]

In the bugzilla issue they didn't fix anything, just closed it and provided a workaround, which in my view is not optimal. Not sure where the change should go, but there should be a /etc/default/lxcfs with a LXCFS_OPTS= var to able to specify whether you want this enabled or not.

I understand they might not want to enforce a behavior change in lxcfs which might not suit everyone, but at least the "workaround" should survive lxcfs upgrades.

 

[tuxillo]

Actually I think I'll send a patch but I have no clue if it'll be accepted.

 

[coldplug]

Even in 8.0 this thing is not fixed.

 

[mailinglists]

I guess the patch was not accepted or sent.

 

[tuxillo]

I guess the patch was not accepted or sent.

Indeed, not yet sent. Will report when I do it

 

[aradabir007]

Still not fixed.

 

[tuxillo]

I've sent a patch to the lists, still waiting for approval to be posted.

Subject: [PATCH lxcfs 0/1] d/rules: Allow passing options to the lxcfs systemd unit
Reason: Post by non-member to a members-only list

Dunno if it's a valid solution since it lacks validations. I did some basic testing in a VM and it seemed to work.

 

[LnxBil]

I've sent a patch to the lists, still waiting for approval to be posted.

Subject: [PATCH lxcfs 0/1] d/rules: Allow passing options to the lxcfs systemd unit
Reason: Post by non-member to a members-only list

Dunno if it's a valid solution since it lacks validations. I did some basic testing in a VM and it seemed to work.

Why don't you become a member? It's just one click.

 

[tuxillo]

The patch went in to the lists but so far no reply: https://lists.proxmox.com/pipermail/pve-devel/2024-May/064114.html

 

[tuxillo]

@fabian any chance this gets reviewed and/or integrated?

 

[fabian]

probably it was missed because its timestamp was almost a month before it got delivered, so it likely sorted quite far back for most recipients I'll take a look..

 

[tuxillo]

probably it was missed because its timestamp was almost a month before it got delivered, so it likely sorted quite far back for most recipients I'll take a look..

Yeah, it took a while to get it accepted. I should have subscribed to the list as mentioned here in the forum... Anyways, thanks for checking! Let me know if I should send the patch upstream instead.

 

[tuxillo]

@fabian did you manage to have a look? thanks

 

[fabian]

see on-list