Bug 740899

Looks like both F14 and RHEL 6 added support for files in /etc/sysctl.d/*, so I'm working on a patch to have libvirt install /etc/sysctl.d/libvirtd.sysctl as needed.

Upstream patch awaiting review:
https://www.redhat.com/archives/libvir-list/2011-October/msg00075.html

In POST:
http://post-office.corp.redhat.com/archives/rhvirt-patches/2011-October/msg00195.html

can reproduce on 
qemu-kvm-0.12.1.2-2.195.el6.x86_64
kernel-2.6.32-206.el6.x86_64
libvirt-0.9.4-12.el6.x86_64

test on machine intel-e7420-128-1
change the  /proc/sys/fs/aio-max-nr to 65535

with attach 513 disks like
    <disk type='block' device='disk'>
      <driver name='qemu' type='raw' cache='none' io='native'/>
      <source dev='/dev/vg0/guest1-u'/>
      <target dev='vdu' bus='virtio'/>
    </disk>

when start 21th guests, it reports error 

error: Failed to start domain guest21
error: internal error Process exited while reading console log output: char device redirected to /dev/pts/22
qemu-kvm: -drive file=/dev/vg0/guest21-l,if=none,id=drive-virtio-disk11,format=raw,cache=none,aio=native: could not open disk image /dev/vg0/guest21-l: Invalid argument

on libvirt-0.9.4-16.el6.x86_64, report the same error, qemu-kvm bug is still in New status, so add depends on bug 739489

Remember that the libvirt fix is to set a sysfs setting at boot.  The only way to verify the behavior of this fix is to install the desired libvirt (whether to prove the bug exists in older libvirt or to prove it is fixed with newer libvirt), then reboot the system, then 'cat /proc/sys/fs/aio-max-nr', then run the test with more than 256 disks.  The reboot between each change of libvirt is essential to prove that /etc/sysctl.d is being sourced correctly at bootup as part of the fix.

(In reply to comment #23)
> Remember that the libvirt fix is to set a sysfs setting at boot.  The only way
> to verify the behavior of this fix is to install the desired libvirt (whether
> to prove the bug exists in older libvirt or to prove it is fixed with newer
> libvirt), then reboot the system, then 'cat /proc/sys/fs/aio-max-nr', then run
> the test with more than 256 disks.  The reboot between each change of libvirt
> is essential to prove that /etc/sysctl.d is being sourced correctly at bootup
> as part of the fix.

Thanks for Eric's reminding. I retest with the following steps on libvirt-0.9.4-16

1. install libvirt-0.9.4.12 and reboot host 
2. check the aio limit
#cat /proc/sys/fs/aio-max-nr 
65536
3. Start 22 guests with 24 disks on each guest and aio=native
On the 22th guest start, it will report error like
error: Failed to start domain guest22
error: internal error Process exited while reading console log output: char device redirected to /dev/pts/23
qemu-kvm: -drive file=/var/lib/libvirt/images/guest22-h.img,if=none,id=drive-virtio-disk7,format=raw,cache=none,aio=native: could not open disk image /var/lib/libvirt/images/guest22-h.img: Inappropriate ioctl for device
4. update to libvirt-0.9.4-16 and reboot host again
5. check the aio limit again
#cat /proc/sys/fs/aio-max-nr
1048576
6. Start 22 guests with 24 disks on each guest and aio=native
All the guest can started successfully

So verify pass 

qemu-kvm-0.12.1.2-2.195.el6.x86_64
kernel-2.6.32-206.el6.x86_64

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

http://rhn.redhat.com/errata/RHBA-2011-1513.html