[kata-dev] virtio-fs + VM Templating
ganesh.mahalingam at intel.com
Wed Jun 5 15:53:20 UTC 2019
> -----Original Message-----
> From: Peng Tao [mailto:bergwolf at hyper.sh]
> Sent: Wednesday, June 5, 2019 12:06 AM
> To: Dr. David Alan Gilbert <dgilbert at redhat.com>; Mahalingam, Ganesh
> <ganesh.mahalingam at intel.com>
> Cc: kata-dev at lists.katacontainers.io
> Subject: Re: [kata-dev] virtio-fs + VM Templating
> On Tue, Jun 4, 2019 at 5:53 PM Dr. David Alan Gilbert <dgilbert at redhat.com>
> > * Tao Peng (bergwolf at hyper.sh) wrote:
> > > On Tue, Jun 4, 2019 at 7:18 AM Mahalingam, Ganesh
> > > <ganesh.mahalingam at intel.com> wrote:
> > > >
> > > > Peng, Stefan,
> > > > While working on enabling virtio-fs on Kata there was a small
> confusion on whether virtio-fs can work with VM Templating. With the current
> setup of kata, VM templating and virtio-fs are mutually exclusive (VM
> templating is enabled in qemu-lite and virtio-fs currently only with NEMU),
> but they will converge very soon.
> > > >
> > > > An open question we had was, would they both work with each other
> given that VM templating turns off memory sharing for all VMs other than the
> first one and virtio-fs expects guest memory be shared & file-backed for all
> > > Hi Stefan,
> > >
> > > What happens if we specify share=on and share=off for different
> > > memory file backends? Can virtio-fs only pick up the shared one?
> > All vhost-user devices (including virtio-fs) need the share=on -
> > because that's the only way that the vhost-user process sees the
> > writes done by qemu or the guest. I think we need it for all of RAM.
> Hmm, vm templating needs the initial ram to be mapped private so that it can
> be shared safely across multiple VM instances. It does conflict with vhost-
> user requirement that guest memory can be accessed by a different process.
> Ganesh, how about we just marking vm templating and vhost-user conflict with
> each other?
Currently that is what we have done. If you enable VM-templating virtio-fs cannot be and scheduling of the pod/container should fail. Will check the code to confirm.
> Into something rich and strange.
More information about the kata-dev