[libvirt-users] Qemu backend: Cannot start new domains after some time

[Guido Winkelmann]

Hi,

After using libvirt+qemu for a while, I'm finding that I can no longer 
reliably start new domains. Most of the time when I try, I'm getting this, or 
something like it:

virsh # start dhcpv6server
error: Failed to start domain dhcpv6server
error: monitor socket did not show up.: No such file or directory

The same thing happens when trying to start the domain using the C-API.

This happens particularly often when there are any domains running that are 
using a lot of CPU, even when most of the physical cores of the host machine 
are still sitting idle.

(I've created a number of Linux VMs, with one CPU core each, that do nothing 
but calculate primes in an endless loop for load-testing purposes.)

If I destroy all of the high-load domains, I can usually start up new domains 
again.

The time between requesting the domain startup and the error message being 
printed varies greatly, from less than a second to more than a minute.

The host machine is running Fedora 12 with libvirt 0.8.1 and qemu-kvm 0.12.4.

Any ideas what might be causing this, and what can be done about it? It would 
be a big problem for us if we cannot start any new domains just because some 
of the existing ones are using a lot of CPU...

Regards,

	Guido