Open Stack

On Fri, 2023-09-01 at 14:56 +0200, Karera Tony wrote:
> Hello Danny,
> 
> Thanks for the feedback.
> 
> use a lowest common denominator cpu_model : Does this mean changing the
> servers ?

it means tha tif you have a mix of cpus in the deployment you shoudl hardcod
to the older of the diffent cpu models i.e skylake or whatever it may be in your case.

https://docs.openstack.org/nova/latest/configuration/config.html#libvirt.cpu_models

[libvirt]
cpu_mode=custom
cpu_models=skylake

in more recent release we replaced the older cpu_model with cpu_models with is a comma sperated
list of models in prefernce order. i.e. nova will use the first cpu_modle in the list that work
on the current host. not that while this give extra flexiblity it vs just using the oldest cpu model
it does limit the set of hosts you can migrate too but means you can have better perfromance so its
a tradeoff. performance vs portablity.

the error you mentioned can also be caused by micorocode updates.
intel remove TSX and MPX i beilve in the last 2 years form come of there cpus
that breaks live migration if a vm was create with access to that cpu instruction

the only way to resolve that is to cold migrate the guest.
i.e. if a vm is currently booted with cpu_model X it cannot be modifed while the guest is running
so you either need to update the config option and reboot all the vms or more particlaly update the config
and then cold migrate them which will allow you to move the instnace(your orginal goal) while allowing
the new cpu model to take effect.

novas default for the cpu model when not set is with our default cpu_mode is host_model
meaning whatever model best matches the current hosts cpu.

that is the correct default in general but if you have mixed cpu generatiosn in your cloud its not ideal.

hopefuly that helps

> 
> Regards
> 
> Tony Karera
> 
> 
> 
> 
> On Fri, Sep 1, 2023 at 12:37 PM Danny Webb <Danny.Webb at thehutgroup.com>
> wrote:
> 
> > It means that you have CPUs with incompatible flags or you've got
> > differing kernel versions that expose different flags or you've got
> > differing libvirt versions that expose different flags.  You either need to
> > use a lowest common denominator cpu_model or do a cold migration.
> > ------------------------------
> > *From:* Karera Tony <tonykarera at gmail.com>
> > *Sent:* 01 September 2023 10:13
> > *To:* openstack-discuss <openstack-discuss at lists.openstack.org>
> > *Subject:* Live migration error
> > 
> > 
> > * CAUTION: This email originates from outside THG *
> > ------------------------------
> > Hello Team,
> > 
> > I am trying to migrate instances from one host to another but I am getting
> > this error.
> > 
> > *Error: *Failed to live migrate instance to host "compute1". Details
> > <http://10.10.13.31/admin/instances/#message_details>
> > Migration pre-check error: Unacceptable CPU info: CPU doesn't have
> > compatibility. 0 Refer to
> > http://libvirt.org/html/libvirt-libvirt-host.html#virCPUCompareResult
> > (HTTP 400) (Request-ID: req-9dfdfe2e-70fa-481a-bd3d-c2c76a6e93da)
> > 
> > Any idea on how to fix this?
> > 
> > Regards
> > 
> > Tony Karera
> > 
> > 
> >