Xen users - Jan 2012 - Xen Ubuntu Pangolin HVMs not recreating properly

I am creating Ubuntu pangolin Hvms (kernel 3.2.0-2-generic) with pvonhvm 
drivers enabled. The host box runs debian squeeze.

The hvms work perfectly when created using the console (-c) option,

''xm create vm_name.cfg -c''

followed by a boot from vncviewer to the host machine. Once I select the 
kernel I want in the vncviewer screen and enable the boot, subsequent 
logins from this or separate shells work perfectly.

But if I shut down or destroy this hvm instance, and restart it without 
the console (-c) option,

I sometimes cannot login and consistently get read-only file system 
errors when attempting to execute commands via sudo.


What could be the problem here?

I find I can do an xm save  / xm restore properly, but not either
xm shutdown   /  xm create
or
xm destroy   /   xm create
or
sudo reboot (invoked from within the running vm)

After any of these 3 I cannot login.

So perhaps this is a boot problem?


On 04/01/12 18:11, Benjamin Weaver wrote:
> I am creating Ubuntu pangolin Hvms (kernel 3.2.0-2-generic) with pvonhvm
> drivers enabled. The host box runs debian squeeze.
>
> The hvms work perfectly when created using the console (-c) option,
>
> ''xm create vm_name.cfg -c''
>
> followed by a boot from vncviewer to the host machine. Once I select the
> kernel I want in the vncviewer screen and enable the boot, subsequent
> logins from this or separate shells work perfectly.
>
> But if I shut down or destroy this hvm instance, and restart it without
> the console (-c) option,
>
> I sometimes cannot login and consistently get read-only file system
> errors when attempting to execute commands via sudo.
>
>
> What could be the problem here?
>
>
>
>

Figured out a solution. The problem does not lie with Xen, but with 
recent releases of Ubuntu.

It was indeed a boot problem. The grub timeout functionality enabling 
automatic boot from grub menu item 0, does not always work in trecent 
Ubuntu versions. This owes to the use by these versions of a recordfail 
parameter. The best solution is claimed by several users to be simply to 
comment out the recordfail code in /etc/grub.d/00_Header:

...
#if [ "\${recordfail}" = 1 ]; then
#  set timeout=-1
#else
   set timeout=${2}
#fi
...


and then to run update-grub


Thus the HVM boots up successfully when using xm create





On 05/01/12 15:09, Benjamin Weaver wrote:
> I find I can do an xm save  / xm restore properly, but not either
> xm shutdown   /  xm create
> or
> xm destroy   /   xm create
> or
> sudo reboot (invoked from within the running vm)
>
> After any of these 3 I cannot login.
>
> So perhaps this is a boot problem?
>
>
> On 04/01/12 18:11, Benjamin Weaver wrote:
>> I am creating Ubuntu pangolin Hvms (kernel 3.2.0-2-generic) with
pvonhvm
>> drivers enabled. The host box runs debian squeeze.
>>
>> The hvms work perfectly when created using the console (-c) option,
>>
>> ''xm create vm_name.cfg -c''
>>
>> followed by a boot from vncviewer to the host machine. Once I select
the
>> kernel I want in the vncviewer screen and enable the boot, subsequent
>> logins from this or separate shells work perfectly.
>>
>> But if I shut down or destroy this hvm instance, and restart it without
>> the console (-c) option,
>>
>> I sometimes cannot login and consistently get read-only file system
>> errors when attempting to execute commands via sudo.
>>
>>
>> What could be the problem here?
>>
>>
>>
>>