[fedora-virt] possible memory leak ?

[Dor Laor]

On 04/20/2010 05:17 PM, Razvan Radu wrote:
> hello,
>
> after running an stress program in the guest for 6 to 24 hours the
> qemu-kvm process uses up all the available memory and gets killed by
> oom, is this an bug or an configuration problem ?
>
> host:
> arch: intel i7 12gb ram
> fedora 12 updated as of 2010-04-19 with fedora-virt packages
> kernel: 2.6.32.11-99.fc12.x86_64
> qemu: qemu*0.12.3-7.fc12.x86_64, also tested with qemu*0.12.3-6.fc12.x86_64
> libvirt: libvirt*0.7.7-2.fc12.x86_64
> qemu command: /usr/bin/qemu-kvm -S -M pc-0.11 -enable-kvm -m 3072 -smp
> 4,sockets=4,cores=1,threads=1 -name guest-00574 -uuid
> 3f26c499-b029-3e1c-2410-fd197b262057 -nodefaults -chardev
> socket,id=monitor,path=/var/lib/libvirt/qemu/guest-00574.monitor,server,nowait
> -mon chardev=monitor,mode=readline -rtc base=utc -boot dc -drive
> if=none,media=cdrom,id=drive-ide0-1-0 -device
> ide-drive,bus=ide.1,unit=0,drive=drive-ide0-1-0,id=ide0-1-0 -drive
> file=/dev/hdd.img/guest-00574-1,if=none,id=drive-virtio-disk0,boot=on,cache=none
> -device
> virtio-blk-pci,bus=pci.0,addr=0x4,drive=drive-virtio-disk0,id=virtio-disk0
> -device
> virtio-net-pci,vlan=0,id=net0,mac=54:52:00:00:00:49,bus=pci.0,addr=0x5
> -net tap,fd=21,vlan=0,name=hostnet0 -usb -device usb-tablet,id=input0
> -vnc 0.0.0.0:2 -vga cirrus -device
> virtio-balloon-pci,id=balloon0,bus=pci.0,addr=0x3
>
> guest:
> CentOS release 5.4
> kernel: 2.6.18-164.15.1.el5
>
> stress program:
> http://weather.ou.edu/~apw/projects/stress/
> comman line: ./stress -v -c 4 -m 4 --vm-bytes 800M -d 4 --hdd-bytes 100M
> -t 100000
>
>
> the qemu process is constantly using more and more memory and after a
> while it get killed
> I have tested with:
> - no hugepages
> - hugepages
> - lvm
> - qcow2 image files
> - cache=writeback
> - cache=none
> all resulted in the same memory usage pattern (slowly growing)
>
>
> do you know of an formula for calculating the memory overhead for
> qemu-kvm (without an potential memory leak ...) ?
> I have noticed an 20% overhead with and without hugepages (pmap reported
> a total of 1.2G for an 1G guest and 3.6G total for an 3G guest)
> is 20% an correct assumption (it seems a little high) ?
> I have observed the same overhead with hugepages, is this normal ?

The is ruffly looks fine, in addition to the guest's ram there is 
overhead for pci config space of the devices, qcow2 metadata, etc.

How much ram do you have on your host? Please provide the top data, 
vmstat 1 and slabtop info.

qemu do copy IO requests but eventually each request should be freed.

>
>
> thanks,
>