libvirt users - Jul 2017 - Re: Changing <on_reboot> in the domain XML file

Using virsh edit, I made the following change to the domain XML file:

<on_reboot>destroy</on_reboot>

But this appears to do nothing. The VM guest is Windows 10, and when  
directing it to reboot, it still reboots, instead of shutting off the VM.

I was trying to work around some kind of a bug, somewhere, that started  
happening after updating to Fedora 26 and qemu 2.9, where a reboot throws my  
Windows 10 guests into some kind of a bizarre automatic recovery mode, which  
then claims that the hard drive is hosed. It's not, and a forced shutdown  
followed by a cold start boots everything back up like nothing has happened  
(which I didn't figure out until reinstalling one of the guests, sigh...),  
and everything is peachy once again.

So, anyway, I was trying to figure out a way around it, and  
<on_reboot>destroy</on-reboot> seemed to be exactly what I was
looking for.

Perusing qemu's man page, it seems that this option should result in a -no- 
reboot option getting added to qemu's command line. But, looking at the  
actual command line, after starting the VM, it's nowhere to be seen.

A bit more Google-fu found this:

http://blog.vmsplice.net/2011/04/how-to-pass-qemu-command-line-options.html

And I manually added a -no-reboot option to the domain XML file that way.

And you know what? After doing that, telling Windows 10 to reboot simply  
shuts off the VM. Wonderful.

But that still leaves me wondering: what's up with the <on_reboot>
tag? Why
didn't it work? Is this a PEBCAK, or something worth throwing a ticket for?

On 07/30/2017 03:08 AM, Sam Varshavchik wrote:
> Using virsh edit, I made the following change to the domain XML file:
> 
> <on_reboot>destroy</on_reboot>
> 
> But this appears to do nothing. The VM guest is Windows 10, and when
> directing it to reboot, it still reboots, instead of shutting off the VM.
> 
> I was trying to work around some kind of a bug, somewhere, that started
> happening after updating to Fedora 26 and qemu 2.9, where a reboot
> throws my Windows 10 guests into some kind of a bizarre automatic
> recovery mode, which then claims that the hard drive is hosed. It's
not,
> and a forced shutdown followed by a cold start boots everything back up
> like nothing has happened (which I didn't figure out until reinstalling
> one of the guests, sigh...), and everything is peachy once again.
> 
> So, anyway, I was trying to figure out a way around it, and
> <on_reboot>destroy</on-reboot> seemed to be exactly what I was
looking for.
> 
> Perusing qemu's man page, it seems that this option should result in a
> -no-reboot option getting added to qemu's command line. But, looking at
> the actual command line, after starting the VM, it's nowhere to be
seen.
> 
> A bit more Google-fu found this:
> 
> http://blog.vmsplice.net/2011/04/how-to-pass-qemu-command-line-options.html
> 
> And I manually added a -no-reboot option to the domain XML file that way.
> 
> And you know what? After doing that, telling Windows 10 to reboot simply
> shuts off the VM. Wonderful.
Libvirt doesn't append -no-reboot because of this commit:

libvirt.org/git/?p=libvirt.git;a=commitdiff;h=0e034efaf9

Having said that, I believe libvirt puts -no-shutdown on your command
line. Then, libvirt has a handler for various guest events - e.g. when
the guest reboots (well, when it calls 'reset'). But when looking at its
implementation it looks like there is something missing:

http://libvirt.org/git/?p=libvirt.git;a=blob;f=src/qemu/qemu_process.c;h=a11d75b80fb018da777fa013a1c5c0762568ac19;hb=HEAD#l478


Compare it to say SHUTDOWN event:

http://libvirt.org/git/?p=libvirt.git;a=blob;f=src/qemu/qemu_process.c;h=a11d75b80fb018da777fa013a1c5c0762568ac19;hb=HEAD#l631

Therefore I think this is a bug. Let me see if I can propose a fix for
it. Meanwhile - you can report a bug if you want.

Michal

Michal Privoznik writes:
> Libvirt doesn't append -no-reboot because of this commit:
>
> libvirt.org/git/?p=libvirt.git;a=commitdiff;h=0e034efaf9
>
> Having said that, I believe libvirt puts -no-shutdown on your command
Yes, I see that option getting passed in.
> line. Then, libvirt has a handler for various guest events - e.g. when
> the guest reboots (well, when it calls 'reset'). But when looking
at its
> implementation it looks like there is something missing:
The end result is that left to its own devices, the guest still reboots,  
with -no-shutdown.
> Therefore I think this is a bug. Let me see if I can propose a fix for
> it. Meanwhile - you can report a bug if you want.
Ok, it's bug 1476866