Running mixed FC5/FC6 domains

[Andy Burns]

Now that FC6T1 is upon us, I'm wondering if it is supported to run an 
FC6Tx domU as a guest of an FC5 dom0 host? Or an FC5 domU as a guest of 
an FC6 dom0 host? If possible would this require synchronised xen 
versions for guest and host domains?

I see using xen as a good way to make upgrades less painful by live 
migration of guests between hosts while the hosts are upgraded, and by 
splitting functions of a server into multiple guests which can be 
tested/upgraded individually.

Allowing mixed domains could ease the "upgrade treadmill" of Fedora, and 
reduce the possibility of breaking everything at once during upgrades.

Presumably RHEL5 customers will eventually ask the same questions about 
supporting mixed RHEL6 upgrades with xen? Getting this right would give 
xen/linux the "no downtime" functionality I recall from performing 
rolling upgrades across nodes in VMS clusters years ago.

Thoughts?

[Ian Kent]

On Sat, 2006-06-24 at 18:38 +0100, Andy Burns wrote:
>  Now that FC6T1 is upon us, I'm wondering if it is supported to
> run an 
>  FC6Tx domU as a guest of an FC5 dom0 host? Or an FC5 domU as a guest of 
>  an FC6 dom0 host? If possible would this require synchronised xen 
>  versions for guest and host domains? 
In theory this should be possible but I'm still trying and not able to
do fc6 (guest) on fc5. I've been told that fc5 (guest) on fc6 (actually
just running an fc6 kernel) should work but I've had no success with
that yet either.

Mind you my requirement is a bit of a big ask I guess, in that I would
only consider this to "work" if I can run the xenguest-install.py
successfully.

There is apparently an incompatibility in the kernel interface as fc6
has a later snapshot of xen.

I expect things will improve soon.

>  
>  I see using xen as a good way to make upgrades less painful by live 
>  migration of guests between hosts while the hosts are upgraded, and by 
>  splitting functions of a server into multiple guests which can be 
>  tested/upgraded individually.
>  
>  Allowing mixed domains could ease the "upgrade treadmill" of Fedora, and 
>  reduce the possibility of breaking everything at once during upgrades.
>  
>  Presumably RHEL5 customers will eventually ask the same questions about 
>  supporting mixed RHEL6 upgrades with xen? Getting this right would give 
>  xen/linux the "no downtime" functionality I recall from performing 
>  rolling upgrades across nodes in VMS clusters years ago. 
I expect questions like this will come in think and fast.
Setting up virtual test and development environments for multiple OS
releases is likely considered "bread and butter functionality" for a
product like xen.

Ian

[Roy-Magne Mo]

Ian Kent wrote:
>  On Sat, 2006-06-24 at 18:38 +0100, Andy Burns wrote:
> > Now that FC6T1 is upon us, I'm wondering if it is supported to run an 
> > FC6Tx domU as a guest of an FC5 dom0 host? Or an FC5 domU as a guest of 
> > an FC6 dom0 host? If possible would this require synchronised xen 
> > versions for guest and host domains?
>  
>  In theory this should be possible but I'm still trying and not able to
>  do fc6 (guest) on fc5. I've been told that fc5 (guest) on fc6 (actually
>  just running an fc6 kernel) should work but I've had no success with
>  that yet either.
>  
>  Mind you my requirement is a bit of a big ask I guess, in that I would
>  only consider this to "work" if I can run the xenguest-install.py
>  successfully.
>  
>  There is apparently an incompatibility in the kernel interface as fc6
>  has a later snapshot of xen.
>  
>  I expect things will improve soon. 
I just installed a rawhide installation as a XEN guest today on a
updated FC5 dom0 with xen-3.0.2-3.FC5 and kernel
2.6.17-1.2139_FC5xen0.

But after installing and booting a couple of times I get this error
message:

Error: Device 0 (vif) could not be connected. Hotplug scripts not working.

A boot clears the problem, so I guess it is some kind of resources that
isn't cleaned up, probably related to (virtual) interfaces. All guests
are LVM-backed so it's not loop devices that's exhausted.