Re: [Qemu-devel] [PATCH] Migration compatibility for serial

[Michael S. Tsirkin]

On Wed, Jun 17, 2015 at 06:34:55PM +0200, Juan Quintela wrote:
> "Michael S. Tsirkin" <address@hidden> wrote:
> > On Wed, Jun 17, 2015 at 02:20:42PM +0200, Paolo Bonzini wrote:
> >> 
> >> 
> >> On 17/06/2015 14:07, Dr. David Alan Gilbert wrote:
> >> > Of course for us on RHEL our minor releases don't correspond to
> >> > QEMU minor releases, so we already support migrating from our
> >> > downstream 7.1 (QEMU 2.1) derivative to our 7.0 (1.5.3) version.  
> >> > And the reason for this patch series is to support something >2.2
> >> > migrating back to that 2.1 (or maybe even to that 1.5.3).
> >> > 
> >> > I don't believe we're alone in wanting to be able to do that type
> >> > of thing;
> >> 
> >> Others may prefer to have migration only work when it is absolutely sure
> >> that it works.  It is much easier to add hacks on top of what upstream
> >> QEMU does (e.g. using the static checker), than to remove the hacks.
> >> 
> >> If we really didn't care about others' support for bidirectional
> >> migration, we would have kept the static checker internal to Red Hat.
> >> Or we wouldn't have bothered to refine the .needed functions, and so on.
> >> 
> >> Paolo
> >
> > What we need to decide is how major is the breakage.
> > If it's minor - like some lost characters - then it's not
> > worth breaking migration for most users.
> > And I think this should be a property so people can
> > force strict mode if they really want to.
> >
> > If it's a major breakage, it's harder to decide:
> > some people might be able to retry migration later.
> > Maybe a flag to enable this mode would make sense?
> > Also, maybe it would be better to fail migration on source
> > rather than send something destination can't handle?
> 
> Source don't know if destination understand it or not.  That is the
> whole point of being optional.  Source sends it if it is needed.
> Destination can handle it (or not).
> 
> there are (at least) two qemu pc-2.2:
> qemu-2.2 -M pc-2.2
> qemu-2.3 -M pc-2.2
> 
> Same machine type.  Second is able to receive it.  First one is not.
> Source don't know what is on the other side.  If user is going to put a:
> 
> --dont_send_serial_because_I_don't_care
> 
> Then it can as well just disable the serial device and live with it.
> 
> Later, Juan.

Just losing worst-case a couple of characters is not the same as
losing serial functionality.

We could have a flag to tell us what's on the other side,
but that would need even more testing. So let's keep it
simple.


> 
> >
> > But let's see what the symptoms are before we argue
> > about this option.
> >
> >> > so you can either worry about not burdening upstream
> >> > with compatibility patches like this, or think it's not fair
> >> > to leave them out if others upstream might want them.  How many
> >> > others? Well I'd say it's got to be more than some of the other
> >> > obscure features in QEMU!