Re: [PATCH 2/5] migration/postcopy: magic value for postcopy channel

[Peter Xu]

On Thu, Nov 07, 2024 at 04:57:46PM +0000, Daniel P. Berrangé wrote:
> On Thu, Nov 07, 2024 at 11:17:30AM -0500, Peter Xu wrote:
> > On Thu, Nov 07, 2024 at 12:33:17PM +0000, Daniel P. Berrangé wrote:
> > I'll comment on a few examples above, which I think some of them, even if
> > handshake is ready, may still need mgmt layers to involve..
> > 
> > Multifd and postcopy are the two major features, and they, IMHO, all better
> > need user involvements..
> > 
> > Multifd needs it because it relies on the channel being able to provide >1
> > channels.  It means "| nc XXX > file" can stop working if we turn it on by
> > default silently.
> 
> NB, my point was referring to a hypothetical alternative history,
> where we had the handshake at the QEMU level from day 1. That
> would neccessarily imply a bi-directional channel, so the 'nc'
> use case would already have been  out of scope. That said, QEMU
> could identify whether the channel it was told to use was
> bi-directional or not, and thus not try to do multifd over
> a non-socket transport.

Ah, that's true.

> 
> So the general point still holds - if QEMU had this protocol
> negotiation phase built-in, there would be more flexiblity in
> introducing new features without layers above needing changes,
> for every single one, just a subset.

Yes.

Just to mention, we can already do that now without handshake, as long as
the feature has zero side effect, and as long as we don't expose it as a
migration capability (which by default all off).  In that case, we can make
the property to "on", and add "off" in machine compat properties.  IOW,
machine compat property can play part of the role as handshake, based on
the machine type must be the same when initiating QEMU on both hosts.

Thanks,

-- 
Peter Xu