|Summary:||Cannot create new virtual machines in virt-manager|
|Product:||[Fedora] Fedora||Reporter:||Casey Dahlin <cdahlin>|
|Component:||virt-manager||Assignee:||Daniel Berrange <berrange>|
|Status:||CLOSED DUPLICATE||QA Contact:||Fedora Extras Quality Assurance <extras-qa>|
|Version:||9||CC:||crobinso, hbrock, vanhoof|
|Fixed In Version:||Doc Type:||Bug Fix|
|Doc Text:||Story Points:||---|
|Last Closed:||2008-12-01 19:44:37 UTC||Type:||---|
|oVirt Team:||---||RHEL 7.3 requirements from Atomic Host:|
|Cloudforms Team:||---||Target Upstream Version:|
Description Casey Dahlin 2008-06-26 15:53:25 UTC
After attempting to install a virtual machine, I got an error during the OS installer. I destroyed the VM, deleted the image from /var/lib/libvirt/images, and tried again. Now I can't click finish in the virtmanager create new virtual system wizard without getting a "Domain has already been started!" error message with this backtrace: Unable to complete install '<type 'exceptions.RuntimeError'> Domain has already been started! Traceback (most recent call last): File "/usr/share/virt-manager/virtManager/create.py", line 651, in do_install dom = guest.start_install(False, meter = meter) File "/usr/lib/python2.5/site-packages/virtinst/Guest.py", line 837, in start_install self.validate_parms() File "/usr/lib/python2.5/site-packages/virtinst/FullVirtGuest.py", line 236, in validate_parms Guest.Guest.validate_parms(self) File "/usr/lib/python2.5/site-packages/virtinst/Guest.py", line 951, in validate_parms raise RuntimeError, _("Domain has already been started!") RuntimeError: Domain has already been started! ' I'm trying to use qemu-kvm, with a 4GiB file image (which virtmanager successfully creates) 512MB of ram, 1 vcpu (on a dual core box), default network settings. Relevant versions: Kernel: 188.8.131.52-18.fc9.x86_64 qemu: 0.9.1-5.fc9 libvirt: 0.4.2-3.fc9 virt-manager: 0.5.4-4.fc9 Will see if a reboot solves this. Component is almost certainly wrong, but I figured I'd start it at the top.
Comment 1 Cole Robinson 2008-06-26 16:30:22 UTC
Destroy doesn't actually delete a VM, it would be more accurately named 'Force Poweroff'. Did you destroy the vm via the command line or virt-manager? Also the proper way to delete a VM is either the delete button in the manager window of virt-manager or use 'virsh undefine'. So just to be sure, the first install fails, you destroy the vm then delete the config, the hit finish on the still open install screen?
Comment 2 Casey Dahlin 2008-06-26 17:40:03 UTC
I destroyed the vm, and I deleted the config. Sorry to be unclear. On reboot, however, the config was back, as well as a new config which libvirt had claimed to be unsuccessful in creating previously.
Comment 3 Casey Dahlin 2008-06-26 17:46:18 UTC
Ok, it seems what's happening is this: when I click finish, it is in fact creating the virtual machine, but the new config is not showing up in the list, the vnc window is not coming up, and (most oddly) the wizard window is not being closed. Having lost my train of thought during the image creation, I was thinking there was still another step and clicking finish again, which is where the error happens. So the issue is it isn't properly registering when I create a new vm (except for the first time I do it in a sesssion). Simply closing virt-manager and reopening it makes things show up again.
Comment 4 Cole Robinson 2008-06-26 18:19:49 UTC
Can you attach ~/.virt-manager/virt-manager.log please? Thanks
Comment 5 Casey Dahlin 2008-06-26 18:37:35 UTC
Created attachment 310359 [details] virt-manager.log Here you go.
Comment 6 Cole Robinson 2008-06-26 19:16:13 UTC
Okay, what you are seeing is a result of a couple libvirt bugs, one of which is fixed upstream but not yet in fedora. Basically if you create a machine, destroy and delete it, then create another machine with the exact same name, crap hits the fan. Reason was that there was a leak in the destroy python bindings that caused libvirt reference counting to go screwy, and the unfixed bug being that libvirt's hash entries should key on uuid and not name. The workaround is to change the name of the second guest, or restart virt-manager in between the two attempts. This is only an issue when using destroy and creating guests with identical names with a persistent hypervisor connection (which virt-manager uses) My guess is the new libvirt release will land in fedora soon which should solve this issue, so we can keep this bug open till then and make sure it is indeed fixed.