Re: [PATCH] qga: implement 'guest-get-nvidia-smi' command

["Daniel P. Berrangé" <berrange@redhat.com>]

On Tue, Mar 31, 2026 at 02:07:27PM +0100, João Vilaça wrote:
> Monitoring, on the hypervisor side, of GPU workloads running in guests with
> NVIDIA GPU passthrough (VFIO) or vGPU (NVIDIA GRID).
> Operators who want GPU metrics from a fleet of VMs must either SSH into
> each guest or deploy a separate monitoring agent inside the guest itself
> (e.g. DCGM exporter).


IMHO, directly people to deploy an existing monmitoring agent solution
in the guest is likely the right thing todo.

We don't really want the QEMU guest agent expanding arbitrarily such
that it effectively re-invents what's already offered by general
purpose monitoring agents.

If the problem is that those existing monitoring agents need network
support, then IMHO the answer is likely to enhance the agent to use
vsock, rather than add more monitoring commands to QEMU.

With regards,
Daniel
-- 
|: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
|: https://libvirt.org          ~~          https://entangle-photo.org :|
|: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|