Re: [Qemu-devel] [PATCH 0/3] Add 'inline-fd:' protocol for migration

[Dr. David Alan Gilbert]

* Daniel P. Berrangé (address@hidden) wrote:
> On Fri, Apr 12, 2019 at 12:13:51PM -0500, Eric Blake wrote:
> > On 4/12/19 7:20 AM, Yury Kotov wrote:
> > > Hi,
> > > 
> > > I've added 'inline-fd:' proto to simplify migration to/from fd.
> > > I thought about modifying the existing 'fd:' proto but I'm not sure it's a
> > > good idea to change it's contract.
> > > 
> > > Existing 'fd:' proto works with previously added fd by getfd or add-fd 
> > > commands.
> > > If client doesn't want to work with this fd before or after migration 
> > > then it's
> > > easier to send an fd with the migrate-* command. Also, client shouldn't 
> > > maintain
> > > this fd.
> > 
> > While the sentiment of making it easier by having fewer QMP commands
> > might be worthwhile, I'm worried about whether this scales. Having just
> > a limited set of commands that take an fd over SCM rights, and every
> > other command wired to automagically work with existing named fds,
> > scales a lot easier than having to teach individual commands how to take
> > an fd inline.
> 
> Yeah I tend to agree - we use FD passing extensively across QMP/HMP
> and all these commands are written to interact with "getfd". I think
> it would be a step backwards to introduce a special case with
> migration that doesn't use "getfd".

Yes, agreed - as you spotted passing fd's gets a bit hairy, and hence
it's best to keep it to a few places.

Lets figure out the right way of fixing the double-close you spotted
rather than adding new schemes.

Dave

> Regards,
> Daniel
> -- 
> |: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
> |: https://libvirt.org         -o-            https://fstop138.berrange.com :|
> |: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK