On Thu, Aug 16, 2007 at 01:49:39PM +0100, Daniel P. Berrange wrote:
> On Thu, Aug 16, 2007 at 08:30:08AM +0100, Keir Fraser wrote:
> > On 15/8/07 21:00, "Daniel P. Berrange" <berrange@xxxxxxxxxx> wrote:
> > > The only feature this re-factoring looses is the ability to have the
> > > paravirt
> > > text console persistently logged to a file. QEMU allows us to specify that
> > > a character device is either sent to a PTY, or a file, but not both at the
> > > same time. I'm thinking about ways in which QEMU's character device config
> > > syntax & drivers may be extended to allow use of a PTY & File
> > > concurrently.
> > >
> > > The patches should be applied in order - after each individual patch you
> > > should still have a fully operational system for both FV & PV.
> > Have you tested both HVM and PV guests with save/restore, migration, and
> > *failed* save/restore (e.g., by forcing an error in xc_domain_save.c)? The
> > last needs care because qemu-dm must be kept around until the last moment,
> > in case the save or migration fails.
> I've not tested failed save/restore or migration - just normal save/restore
> for FV & PV. I'll do some further testing of the scenarios you suggest. I
> don't think I should have broken anything in qemu-dm for HVM because in the
> HVM codepaths I merely moved code from one file to another.
Migration was working fine, but there's some complications with recovering
from a failed save attempt. In fact even without my changes it currently
fails being unable to detach the blktap disk, and so the recovery dies with
a 'Device vbd already connected' error. Once I track that problem down I
know what changes I need to make PVFB recovery work
|=- Red Hat, Engineering, Emerging Technologies, Boston. +1 978 392 2496 -=|
|=- Perl modules: http://search.cpan.org/~danberr/ -=|
|=- Projects: http://freshmeat.net/~danielpb/ -=|
|=- GnuPG: 7D3B9505 F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 -=|
Xen-devel mailing list