Re: [Qemu-devel] Several questions about " [PATCH v3 0/5] virtio-9p: hotplug and migration support"

[Greg Kurz]

On Tue, 5 Mar 2019 19:28:37 +0800
wangyan <address@hidden> wrote:

> I am very happy for your reply. 
> 
> I am very interested in your old WIP and want to take a look. 
> 
> Our project still use 9pfs, and it needs to live migrate the
> device while it is being used by the guest, is there any advice
> for us to make it come true or any WIP about this? 
> 

It isn't a trivial task, really. The first thing to address is to be able
to drain outstanding I/O requests. Then, about migration itself, the current
idea was to close files at pre-save so that they get re-open automatically
on the target if migration succeeds or on the source if migration fails.
Then we get to some more gory details: files open with O_EXCL and unlinked
open files...

> 
> On 2019/2/26 17:11, Greg Kurz wrote:
> > On Tue, 26 Feb 2019 11:17:16 +0800
> > wangyan <address@hidden> wrote:
> >  
> >> Dear Kurz,
> >>
> >> Good day.
> >>
> >> This is Yan Wang from china. Glad to contact with you.
> >>  
> > Hi Yan Wang,
> >  
> >> I found your patch "[Qemu-devel] [PATCH v3 0/5] virtio-9p: hotplug and 
> >> migration support" on the internet. Recently, our project need to use the 
> >> feature of 9pfs living migration, and I want to know if you have the plan 
> >> to merge the patch to the master branch of qemu, or do you have any other 
> >> plans about the migration feature?
> >>  
> > AFAIK some of these patches have already been merged.
> >
> > https://git.qemu.org/?p=qemu.git;a=commitdiff;h=4652f1640e029e1f2433fa77ba6af285c7cd923a
> > https://git.qemu.org/?p=qemu.git;a=commitdiff;h=6cecf093735f2e5af7d0e29d957350320044e354
> >
> > These are enough to live-migrate a quiescent virtio-9p device, ie,
> > only if the 9p filesystem is NOT mounted in the guest.
> >
> > If your need is to live migrate the device while it is being used by
> > the guest, then this is a different story. Basically, this requires
> > to quiesce the device so that its state can be captured. Also, some
> > state belongs to the host and maybe hard to capture/migrate. For
> > example, if the guest opens a file, write some data in it, unlinks the
> > file but keeps the fd, it expects to be able to read the same data
> > back, even after migration to some other host system.
> >
> > 9pfs is no longer a priority for anyone in the industry. I have no
> > plan to make this go forward, an not much time to work on it. But
> > if you want to have a look, I guess I can share some old WIP I have
> > somewhere.
> >  
> >> Sincerely,
> >>
> >> Yan Wang
> >>  
> > Cheers,
> >
> > --
> > Greg
> >
> > .
> >  
> 
>