[pve-devel] [PATCH] remove openvz fairscheduler code from fairsched patch

Alexandre DERUMIER aderumier at odiso.com
Wed May 27 05:02:50 CEST 2015


I have done tests with cgroups,

and it's working fine with lxc,

For example, I can setup cpu.shares with echo XXXX > /sys/fs/cgroup/cpu/lxc/vmid/cpu.shares (or set it in lxc config)

and it's correctly weighted when host cpu is at 100%

I can see stats for each cgroups
#systemd-cgtop


But for qemu, It doesnt work, I don't known why

mkdir /sys/fs/cgroup/cpu/qemu/vmid/
echo qemupid > /sys/fs/cgroup/cpu/qemu/vmid/tasks
echo XXX > /sys/fs/cgroup/cpu/qemu/vmid/cpu.shares

and
#systemd-cgtop show 0% for cpu usage of the cgroups.


I don't known if it's because of systemd ?

Libvirt manage directly cgroups for lxc and qemu through systemd-dbus api.


But if it's working with lxc and cgmanager, I think it should work too with qemu or any other process..?..


So,I'm stuck ....


----- Mail original -----
De: "dietmar" <dietmar at proxmox.com>
À: "aderumier" <aderumier at odiso.com>
Cc: "pve-devel" <pve-devel at pve.proxmox.com>
Envoyé: Mardi 26 Mai 2015 18:51:29
Objet: Re: [pve-devel] [PATCH] remove openvz fairscheduler code from fairsched patch

> Yes (I think next week, I was very busy last weeks between ceph bench and 
> proxmox training). 
> 
> I need to check how it's works with systemd and how libvirt is managing it. 
> 
> 
> 
> I need to also finish cloud-init support, I have a working prototype, seem to 
> be promising. 

great! 


More information about the pve-devel mailing list