Re: [Qemu-devel] [RFC] qemu snapshot enchancement

[Wenchao Xia]

于 2013-1-25 15:20, Dietmar Maurer 写道:
> > > > > >       My understanding is internal snapshots is obvious fast in both
> > > > > > deleting and reading, and I have similar questions, Dietmar, could
> > > > > > u tip more how you use this case while 2 snapshot layer exist?
> > > > >
> > > > > To be honest, I don't really understand what you talk about here.
> > > > > There are simply different storage types and file formats. There are
> > > > > several storage types with support for internal snapshots:
> > > > >
> > > > > * traditional qcow2 files on normal file system
> > > > > * newer file systems (zfs, btrfs)
> > > > > * lvm (new dm-thin pools)
> > > > > * RDB (ceph)
> > > > > * sheepdog
> > > > > * iSCSI and nexenta API
> > > > > *...
> > > > >
> > > > > And you simply want to allow an external library to create/remove
> > > > > the
> > > > snapshot for you.
> > > > >
> > > >      I remember you have a patch enable take qcow2 internal snapshot
> > > > for every block device, and then will you use LVM to take another LVM
> > > > snapshot for it? If not how did you treat this qcow2 internal snapshot
> > later?
> > > No. Our VMs can have multiple disks on different storages. So we
> > > either use qmp commands to create internal snapshots (qcow2), or
> > > external tools - that depends on storage type.
> > >
> > > Basically we do:
> > >
> > > 1.) save state
> > > 2.) pause VM
> > > 3.) use qmp commands to create internal snapshots for all disks which
> > > supports that
> > > 4.) use external tools to create snapshots for other disks (nexenta,
> > > btrfs, ...)
> > > 5.) resume the VM
> > >
> > > Is that more clear now?
> >
> >     Thanks for declaration, one more question:
> > 3) will have a big image file, will you back up it or just leave it there, not
> > backup for internal snapshot?
>
> Sorry, I do not understand that question. Why will I have a 'a big image file', 
> and
> what do you mean by 'backup' exactly?

To make data safe, I guess usually there would be two copy of image
data, one is the VM used, one is on another "data restore server".
There are choices about how to store the data:
One is do a full clone of image disk, at each time user want to protect
his data.
One is only do a full lone of image disk at first time, later at each
time, only copy out the delta data, which is obvious better.

internal snapshot increase the image file size, for eg, after internal
snapshot 10G image will become 10.1G, and it seems no way to get the
100M delta data and store it in "data restore server", so you will have
do a clone of 10.1G data again. So I asked whether you have done a
"snapshot data copy to restore server" in the following.
--
Best Regards

Wenchao Xia