[libvirt-users] Create and revert to snapshot fails silently after device update on running domain

[Jasper Spit]

Hi Eric,

I ran into this issue again in a slightly different scenario:

1) create a snapshot on running domain

then for backup purposes I do the following:

2) create an external disk snapshot without metadata, and copy off the 
original file to a backup device
3) do a blockpull operation so the external disk snapshot file now 
becomes the base file
4) delete the original file

Now activating the snapshot in the first step results in the same 
behaviour, no snapshot activation and no error is reported.
I can imagine this is a similar problem, as the disk source files 
change (but this is not a hotplug event right?), so I have also tried:

2) create an external disk snapshot without metadata, and copy off the 
original file to a backup device
3) do a blockpull operation so the external disk snapshot file now 
becomes the base file
4) delete the original file
5) create a new external disk snapshot with the original pre-backup 
filename
6) do a blockpull operation again

In this case everything should be the same as prior to the backup, but 
still the snapshot does not activate. Stopping the domain and then 
activating the snapshot also does not work (it just starts the domain 
instead).

Any idea?

Kind regards,
- Jasper

On Thu, 04 Oct 2012 10:44:27 -0600, Eric Blake wrote:
> On 09/26/2012 04:19 AM, Jasper Spit wrote:
>> Hi list,
>>
>> I'm having an issue with snapshot creation. Scenario:
>>
>> qemu 1.1
>> libvirt 0.9.12
>>
>> I create a domain, and start it. The domain has 1 IDE cdrom device
>> defined (see below).
>> When started, I want to mount an ISO file to it. So I do
>> updateDeviceFlags in libvirt-python or update-device in virsh (both 
>> have
>> the same problem).
>
> Both update approaches go through the same libvirt API, so it the 
> action
> of a live device hot-plug that is messing things up here.
>
>> This works fine, the ISO image becomes available to the domain. Now 
>> I
>> create a snapshot on the still running domain using 
>> snapshotCreateXML in
>> libvirt-python or snapshot-create in virsh. The command returns
>> immediately without error (a normal snapshot takes several seconds 
>> to
>> complete).
>
> That sounds like a bug, where libvirt should have given an error 
> about
> being unable to create a snapshot.
>
>> If I revert to this snapshot this command also returns
>> immediately without error, but the snapshot is not actually reverted 
>> to,
>
> Probably fallout from the first bug - if the snapshot was never 
> created
> in the first place, but libvirt went ahead and updated its metadata 
> to
> claim that the snapshot exists, then reverting will have nothing to
> return to.
>
>> the domain remains running in the same state as if nothing had 
>> happened
>> (I test this by verifying console output and checking if a testfile 
>> is
>> present on the domain). If I do not use the update device commands 
>> prior
>> to creating a snapshot, all is well. If I remove the source from the
>> cdrom device using update device, the snapshots work properly again.
>>
>> Any idea what causes this?
>
> It sounds like qemu's 'savevm' command is not very good about 
> handling
> saves after a hotplug event, and that libvirt isn't properly 
> recognizing
> this as a situation in which it must fail the snapshot.  I'll have to
> try and reproduce the setup to see if I can come up with a libvirt
> patch, but your steps look pretty detailed.  Would you mind opening 
> this
> as a BZ, so it doesn't get lost?
>
>>
>> Steps to reproduce using virsh:
>>
>> virsh # start 4d5d722b-864c-657e-0f39-55d1bafc760e
>> Domain 4d5d722b-864c-657e-0f39-55d1bafc760e started
>>
>> virsh # snapshot-create 4d5d722b-864c-657e-0f39-55d1bafc760e
>> Domain snapshot 1348653920 created
>>
>> virsh # snapshot-revert 4d5d722b-864c-657e-0f39-55d1bafc760e 
>> 1348653920
>>
>> All is good, the snapshot is reverted to properly. Now I update the
>> cdrom device:
>>
>> virsh # update-device 4d5d722b-864c-657e-0f39-55d1bafc760e deb.xml
>> Device updated successfully
>>
>> virsh # snapshot-create 4d5d722b-864c-657e-0f39-55d1bafc760e
>> Domain snapshot 1348654116 created
>>
>> Command returns instantly.
>>
>
>>
>> virsh # dumpxml 4d5d722b-864c-657e-0f39-55d1bafc760e
>> <domain type='kvm' id='135'>
>
>>   <devices>
>>     <emulator>/usr/bin/kvm</emulator>
>>     <disk type='file' device='cdrom'>
>>       <driver name='qemu' type='raw'/>
>>       <source 
>> file='/data/images/debian-live-6.0.4-amd64-standard.iso'/>
>>       <target dev='hdc' bus='ide'/>
>>       <readonly/>
>>       <alias name='ide0-1-0'/>
>>       <address type='drive' controller='0' bus='1' target='0' 
>> unit='0'/>
>>     </disk>
>>     <disk type='file' device='disk'>
>>       <driver name='qemu' type='qcow2'/>
>>       <source 
>> file='/data/domains/2f8baacd-563c-b747-b621-c0ddb4aa84bd'/>
>>       <target dev='vda' bus='virtio'/>
>>       <alias name='virtio-disk0'/>
>>       <address type='pci' domain='0x0000' bus='0x00' slot='0x05'
>> function='0x0'/>
>>     </disk>
>
>> </domain>
>>
>> deb.xml:
>>
>> <disk type='file' device='cdrom'>
>>   <driver name='qemu' type='raw'/>
>>   <source file='/data/images/debian-live-6.0.4-amd64-standard.iso'/>
>>   <target dev='hdc' bus='ide'/>
>> </disk>
>>
>> deb-off.xml:
>>
>> <disk type='file' device='cdrom'>
>>   <driver name='qemu' type='raw'/>
>>   <target dev='hdc' bus='ide'/>
>> </disk>
>>
>> Thanks much,
>
> Hopefully we can get this resolved before the next libvirt release.