cPanel confirmed that they're using sysinfo(2) via Perl to get the loads, and lxc would need the kernel patch discussed here https://github.com/lxc/lxcfs/issues/202
we don't want to carry a delta comparted to the upstream kernel(s) like that - it's a slippery slope that ends up with another openvz patchset that never gets upstreamed. it would need to be something with realistic chances of getting included upstream, and people driving that inclusion.
Here's what I got back from cPanel when I asked them about it:
I didn't bother pushing more past that point. If you feel like contacting them and pushing, maybe open a feature request and link it here? I'm sure others will come...
I just received this message from CPanel:
I don't quite understand what they indicate. I mean, I see what they say needs to be configured, but I don't know where I should do it. Anyone with more experience to help?
Thank you!
Thanks very much. This is great!
As a side note, be aware when you restart lxcfs service all the containers are not able to access /proc filesystem, so it is much better to do this once you have shutted down all the containers and not which all of them running.
Hi all
After having the same problem and fixing it as described in this post, now i have the opposite one; the load inside the container is propagated to the host.
I have a 4 cpu container which launches 70 threads, most of them waiting for cpu to be available. The container is using the 4 cpus 100%, and the load inside the container is between 40 and 50, clearly showing the number of processes waiting.
However i did not expect this value to be passed to the host server, the same way a KVM machine does not pass his load to the host server. Although it is totally correct from the load definition, as the host server is seeeing 50-60 process waiting, the fact is they are deliberately contained as per container resources definition and imho should not affect the main server load.
i assume most of us use the load of the server as an indication of their state, but in this case the load is misleading. Is this any way to hide the lxc load from the host?
I plan to deploy a several containers like this always counting, even oversuscribing cpu usage but i am missing a good indicator then about how the host server is performing
After having the same problem and fixing it as described in this post, now i have the opposite one; the load inside the container is propagated to the host.
I have a 4 cpu container which launches 70 threads, most of them waiting for cpu to be available. The container is using the 4 cpus 100%, and the load inside the container is between 40 and 50, clearly showing the number of processes waiting.
However i did not expect this value to be passed to the host server, the same way a KVM machine does not pass his load to the host server. Although it is totally correct from the load definition, as the host server is seeeing 50-60 process waiting, the fact is they are deliberately contained as per container resources definition and imho should not affect the main server load.
i assume most of us use the load of the server as an indication of their state, but in this case the load is misleading. Is this any way to hide the lxc load from the host?
I plan to deploy a several containers like this always counting, even oversuscribing cpu usage but i am missing a good indicator then about how the host server is performing
After Upgrade to Proxmox 7 the -l Trick (https://bugzilla.proxmox.com/show_bug.cgi?id=1870) dont work anymore.
My lxcfs is started with the -l Parameter:
but the Containers shows the Host Load.
My lxcfs is started with the -l Parameter:
Code:
ps | grep lxcfs
root 1273 0.2 0.0 1372520 4472 ? Ssl 14:15 0:06 /usr/bin/lxcfs -l /var/lib/lxcfs/Did you reboot?
It works with a reboot over here.
cat /lib/systemd/system/lxcfs.service
Code:
[Unit]
Description=FUSE filesystem for LXC
ConditionVirtualization=!container
Before=lxc.service
Documentation=man:lxcfs(1)
[Service]
ExecStart=/usr/bin/lxcfs /var/lib/lxcfs -l
KillMode=process
Restart=on-failure
ExecStopPost=-/bin/fusermount -u /var/lib/lxcfs
Delegate=yes
ExecReload=/bin/kill -USR1 $MAINPID
[Install]
WantedBy=multi-user.targetHost load avg: 1.10 2.01 1.59
PCT load avg: 0.10 0.19 0.10
Last edited:
You are right, my machine shows the same behavior.
Load average seems accurate per container, but CPU shows one of the host CPU load instead of the LXC CPU load.
It did work with the latest 6.4 version for me as well, but doesn't work anymore with 7.
Jup, also would like to see that working. I monitor all my guests using zabbix agents and I basically can only monitor filesystems and network interfaces in LXCs because CPU and RAM metrics show wrong results, so unlike with VMs there is noting warning me when the load is too high or RAM is running out of space.
I know that the staff told us in another thread that monitoring software could read the correct CPU and RAM usage when reading it from the right place, but looks like no monitoring software cares about that.
I know that the staff told us in another thread that monitoring software could read the correct CPU and RAM usage when reading it from the right place, but looks like no monitoring software cares about that.
Last edited: