Sorry guys, Daniel is right. I mixed cputune setting of kvm up with that of lxc.
On 2013-10-08 19:14 , [email protected] wrote: Well, Daniel. If the thing is as you said, I'm wrong. But The fact is with the same cputune xml configure under different version of libvirt, It will lead to different results in cgroup. Details are as following: hzguanqiang@debian1:~$ sudo virsh version Compiled against library: libvir 0.9.12 Using library: libvir 0.9.12 Using API: QEMU 0.9.12 Running hypervisor: QEMU 1.1.2 hzguanqiang@debian1:/data/nova/instances/instance-00000053$ cat libvirt.xml <domain type="kvm"> <uuid>2bc80c41-ed34-4a7e-8645-e9af789b29cb</uuid> <name>instance-00000053</name> <memory>8388608</memory> <vcpu cpuset="4-63">8</vcpu> <os> <type>hvm</type> <boot dev="hd"/> </os> <features> <acpi/> </features> <cputune> <shares>32768</shares> <quota>100000</quota> <period>100000</period> </cputune> <clock offset="utc"> <timer name="pit" tickpolicy="delay"/> <timer name="rtc" tickpolicy="catchup"/> </clock> <cpu mode="host-model" match="exact"/> <devices> <disk type="file" device="disk"> <driver name="qemu" type="qcow2" cache="none"/> <source file="/data/nova/instances/instance-00000053/disk"/> <target bus="virtio" dev="vda"/> </disk> <disk type="file" device="disk"> <driver name="qemu" type="qcow2" cache="none"/> <source file="/data/nova/instances/instance-00000053/disk.local"/> <target bus="virtio" dev="vdb"/> </disk> <interface type="bridge"> <mac address="fa:16:3e:79:04:5c"/> <model type="virtio"/> <source bridge="br100"/> <filterref filter="nova-instance-instance-00000053-fa163e79045c"> <parameter name="IP" value="10.160.161.21"/> <parameter name="DHCPSERVER" value="10.160.160.11"/> </filterref> </interface> <serial type="file"> <source path="/data/nova/instances/instance-00000053/console.log"/> </serial> <serial type="pty"/> <input type="tablet" bus="usb"/> <graphics type="vnc" autoport="yes" keymap="en-us" listen="127.0.0.1"/> </devices> </domain> hzguanqiang@debian1:/data/nova/instances/instance-00000053$ cat /sys/fs/cgroup/cpu/libvirt/qemu/instance-00000053/cpu.cfs_quota_us 800000 hzguanqiang@debian2:~$ vir version Compiled against library: libvirt 1.1.2 Using library: libvirt 1.1.2 Using API: LXC 1.1.2 Running hypervisor: LXC 3.9.6 hzguanqiang@debian2:~$ vir dumpxml instance-0000006c <domain type='lxc' id='10303'> <name>instance-0000006c</name> <uuid>ab43db36-ee30-4223-92e8-7652255cffb7</uuid> <memory unit='KiB'>1048576</memory> <currentMemory unit='KiB'>1048576</currentMemory> <vcpu placement='static' cpuset='4-23'>3</vcpu> <cputune> <shares>6144</shares> <period>100000</period> <quota>57499</quota> </cputune> <resource> <partition>/machine</partition> </resource> <os> <type arch='x86_64'>exe</type> <init>/sbin/init</init> <cmdline>console=tty0 console=ttyS0</cmdline> </os> <clock offset='utc'/> <on_poweroff>destroy</on_poweroff> <on_reboot>restart</on_reboot> <on_crash>destroy</on_crash> <devices> <emulator>/usr/libexec/libvirt_lxc</emulator> <filesystem type='mount' accessmode='passthrough'> <source dir='/opt/stack/data/nova/instances/ab43db36-ee30-4223-92e8-7652255cffb7/rootfs'/> <target dir='/'/> </filesystem> <interface type='bridge'> <mac address='fa:16:3e:16:41:47'/> <source bridge='br100'/> <target dev='veth0'/> <filterref filter='nova-instance-instance-0000006c-fa163e164147'/> </interface> <console type='pty' tty='/dev/pts/0'> <source path='/dev/pts/0'/> <target type='lxc' port='0'/> <alias name='console0'/> </console> </devices> <seclabel type='none'/> </domain> hzguanqiang@debian2:~$ cat /sys/fs/cgroup/cpu/machine/instance-0000006c.libvirt-lxc/cpu.cfs_quota_us 57499 >From the above, you can see with libvirt version of 0.9.12, the >'cpu.cfs_quota_us' parameter of instance cgroup is the value of cputune.period multiplied by vcpu numbers. And with libvirt version of 1.1.2, they are equal. On 2013-10-08 17:58 , Daniel Berrange wrote: I don't believe this is correct. In libvirt < 0.10.0 we created a single cgroup $ROOT/libvirt/qemu/$GUESTNAME and set the 'quota * nvcpus' value at that level In libvirt >= 0.10.0 we create one cgroup per vCPU $ROOT/libvirt/qemu/$GUESTNAME/{vcpu0,vcpu1,....} and set the 'quota' value at the vcpuN level. The end result is the same in both cases - you get 'quota * nvcpus' worth of time allowed. -- You received this bug notification because you are subscribed to the bug report. https://bugs.launchpad.net/bugs/1230050 Title: setting of cputune.quota need to recoginse libvirt version Status in OpenStack Compute (Nova): In Progress Bug description: Setting of cputune.quota has changed in libvirt from version 0.10.0. So the cputune.quota parameter of cpu qos which is writed into the libvirt xml should be set differently under different version of libvirt. To manage notifications about this bug go to: https://bugs.launchpad.net/nova/+bug/1230050/+subscriptions ------------------ Best regards! GuanQiang 19:04:26 ------------------ Best regards! GuanQiang 15:12:33 ** Changed in: nova Status: In Progress => Invalid -- You received this bug notification because you are a member of Yahoo! Engineering Team, which is subscribed to OpenStack Compute (nova). https://bugs.launchpad.net/bugs/1230050 Title: setting of cputune.quota need to recoginse libvirt version Status in OpenStack Compute (Nova): Invalid Bug description: Setting of cputune.quota has changed in libvirt from version 0.10.0. So the cputune.quota parameter of cpu qos which is writed into the libvirt xml should be set differently under different version of libvirt. To manage notifications about this bug go to: https://bugs.launchpad.net/nova/+bug/1230050/+subscriptions -- Mailing list: https://launchpad.net/~yahoo-eng-team Post to : [email protected] Unsubscribe : https://launchpad.net/~yahoo-eng-team More help : https://help.launchpad.net/ListHelp
