On 02/13/2012 03:49 PM, Paul Heinlein wrote:

On Mon, 13 Feb 2012, Emmett Culley wrote:

...

Since switching from CentOS 5.7 to CentOS 6.x I find that one of the VMs will not restart upon system reboot. I finally figured out that the video is always set to vmvga after rebooting. In addition the VM's storage always reverts back to IDE from virtio and is set to use the wrong image.

It may be related to the fact that I moved the XML files that defined the VMs from the old (CentOS 5.7) /etc/libvirt/qemu directory to the new (CentOS 6.2) /etc/libvirt/qemu directory. The images are stored is the same LVM partitions as they were on the CentOS 5.7

Any ideas why a VM would revert back? I've checked the XML file before rebooting and it always has the correct info in it, until after the reboot.

Did you copy the XML file from the 5.7 host to 6.3 host as-is or did you edit it at all during the migration?

I've found that I need either to run "virsh define $DOM.xml" and then start the new domain or run "virsh create $DOM.xml" and then, once it's running, make a trivial change (usually adding an XML comment) to the configuration via "virsh edit $DOM" to get things to stick.

I did copy the XML files to the 6.2 machine. Then when I found that the VMs would not start because they were set to vmvga video, which doesn't seem to be supported anymore, I changed the XML files to cirrus.

Then I tried editing the VM using virsh, but still it doesn't persist.

Hmm. There is only one of the VMs that doesn't persist so I'll try using virsh edit again on that one and make a change as you suggest, just in case I never really changed anything for that VM using virsh edit.

Thanks, Emmett

On Mon, 13 Feb 2012, Emmett Culley wrote:

Still doesn't persist. Each time I reboot I have to use virt-manager to change video to cirrus from vmvga, then remove the IDE driver that points to the wrong storage location and add a new virtio storage device pointing to the correct image (an LVM partiiton).

After I make the changes I close virt-manager and restart it, then look at the configuration for the non-persistent VM, and my changes are still there and I can run the VM.

I did a "grep vmvga" on the entire /etc/libvirt directory tree and found no references to "vmvga". Where can libvirt be getting info to change the xml to vmvga, or the IDE to the wrong location?

I've not seen these symptoms, so I can only outline the sorts of steps I'd take in troubleshooting. First, I'd check for SELinux issues:

1. Run "fixfiles check /etc" to see if a configuration file is mislabled.

2. Do the same thing on /var to see if any runtime files have issues.

3. Run "ausearch -m avc" and grep for qemu or libvirt issues.

After that, I'd get more drastic:

1. virsh shutdown $DOM. 2. virsh dumpxml $DOM > /tmp/dom.xml 3. virsh undefine $DOM 4. virsh create /tmp/dom.xml 5. virsh edit $DOM 6. virsh start $DOM --console

And then see if things get better.