On Tue, 2010-01-26 at 17:32 -0500, Anthony Xu wrote:
> On Tue, 2010-01-26 at 14:18 -0800, Daniel Stodden wrote:
> > On Tue, 2010-01-26 at 17:07 -0500, Anthony Xu wrote:
> > > It is clear now, thanks.
> > >
> > > The other thing I'd like to do is how XCP handle disk cache inside VM
> > > when creating a snapshot? I saw from Xencenter seem the VM is stopped
> > > temporarily when creating a snapshot.
> > >
> > > Does VM flush dirty disk cache when creating snapshot?
> > Depends what you mean by disk caches. All I/O performed by the backend
> > non-buffered, so there's presently no need to flush. As soon as a guest
> > I/O request is processed, it essentially goes directly to the disk.
> > The snapshot is created while the VBD is paused, i.e. guest accesses
> > which haven't been issued to the disk are suspended. Next, request which
> > have been sent to the disk are waited for, up to completion. Then blktap
> > closes the handle to the physical disk node.
> > Before resuming guest access, we then reopen the newly created snapshot
> > node, as the new leaf node.
> That means if guest linux is executing "yum install kernel" when
> creating snapshot, the vm created from this snapshot might be not
Yeah, sorry. I got you wrong regarding the VM flush, I thought you were
talking about the backend side. Indeed, we presently don't quiesce Linux
xen-api mailing list