Re: [Qemu-devel] [RFC]VM live snapshot proposal

[Paolo Bonzini]

Il 05/03/2014 02:00, Huangpeng (Peter) ha scritto:
> > > One question:
> > > Can post-copy fallback if exceptions happen during post-copy?
> >
> > What do you mean by 'exceptions' here?  Generally postcopy can't fall back to
> > precopy because once you're in postcopy mode the state is split between the
> > two machines.
>
> Like destination VM interrupted due to memory-copy error or other exceptions,
> with pre-copy scheme, we can fall-back to the source-vm.

No, postcopy cannot do that.

However, this is a limitation of postcopy, not of the kernel interfaces 
that Andrea is adding.  If you use those interfaces to implement live VM 
point-in-time snapshots, you can drop the snapshotting operation safely 
and keep the VM running.

> One simple question(may be discussed before), what kind of scenario does 
> post-copy
> aim for?

Mostly cases where pre-copy migration doesn't converge because the guest 
is too big, or when you require a really, really small downtime.

It can be useful when you have to evacuate a host as fast as possible 
(due to detecting an intrusion or impending hardware failure), because 
the alternative is to shutdown the VM immediately.  It can also be used 
for upgrading QEMU on the host where the VM is running; in this case you 
can use a Unix socket for transport, and eliminate the chance of 
migration failing due to a network problem.

Paolo