Re: [Xen-users] Fw: coordinating VM snapshot with LVM snapshot.

[Pasi Kärkkäinen <pasik@xxxxxx>]

On Thu, Jan 14, 2010 at 10:34:30AM -0800, J Nb wrote:
> I will look in to XCP.. but my guess is that they use their pv drivers to 
> communicate via xenstore..etc. I am looking for something that would work 
> with vanilla open source xen. (well non XCP  :) )
>

Yeah.. most probably.

> But my plan is same as I described in skeleton and organization very similar 
> to yours. but to have something that works with ssh/wmi. This way scripts etc 
> can be centrally managed.
> 

Indeed.

> What I was looking for was 
>  -- For xen is there a flush command for PV and HVM vms. (hypervisor level)
>

Not that I know of. If you use phy: backend for LVM volumes, then dom0
doesn't do any caching.. then it's enough to flush the caches and disks in
the guest.

>  -- Is sync sufficient as a starting point at OS level.
>

I'm not sure.. Linux kernel v2.6.29 added filesystem "freeze" feature
(http://lwn.net/Articles/287435/) that can possibly help here.

I remember reading that "sync" might not really flush everything.. but
yeah, I guess it's good enough as a starting point.

>  -- Any repository of application specific scripts. (Mysql,oracle..etc)
>

Do you mean a website with scripts to share with others? Ideally
application specific scripts would be packaged together with the apps,
assuming this kind of "vssd" was included as a default in the distros :)

>  -- And anything to control cpu/io consumption of the backup process. 
> (Probably start with cstream / nice)
>

This pretty much depends where you do the snapshots and from where you run
the backups.. and also on the used storage solution.

-- Pasi

> -nb
> 
> --- On Thu, 1/14/10, Pasi Kärkkäinen <pasik@xxxxxx> wrote:
> 
> > From: Pasi Kärkkäinen <pasik@xxxxxx>
> > Subject: Re: [Xen-users] Fw: coordinating VM snapshot with LVM snapshot.
> > To: "J Nb" <j_nwb@xxxxxxxxx>
> > Cc: "xen-users" <xen-users@xxxxxxxxxxxxxxxxxxx>
> > Date: Thursday, January 14, 2010, 9:15 AM
> > On Wed, Jan 13, 2010 at 09:11:47AM
> > +0200, Pasi Kärkkäinen wrote:
> > > On Tue, Jan 12, 2010 at 05:44:18PM -0800, J Nb wrote:
> > > > Anyone ?
> > > > 
> > > 
> > > Now that XenServer is opensourced as XCP (Xen Cloud
> > Platform), you could
> > > take a look at the source.. 
> > > 
> > > I believe XenServer supports Windows VSS coordination
> > for snapshots.
> > > "xe vm-snapshot-with-quiesce"
> > > 
> > 
> > Did you get anywhere with this? 
> > 
> > For a long time I've been thinking of writing some kind of
> > 
> > "VSS" library/daemon for Linux.. similar to Microsoft
> > Windows 
> > Volume Snapshot Services.
> > 
> > It could (and should) be called something different than
> > VSS, 
> > but I'll just use that name for now..
> > 
> > The interface/usage could be something like:
> > 
> >     1. backup begins
> >     2. execute in the guest: "vss_control
> > pre-snapshot xvda"
> >     3. take snapshot of the lvm volume
> > having guest disk xvda
> >     4. execute in the guest: "vss_control
> > post-snapshot xvda"
> >     5. do the actual backup from the lvm
> > snapshot
> >     6. backup ends
> > 
> > This would have minimal downtime for the applications
> > running in the guest.
> > 
> > pre-snapshot action would go through all the application
> > specific pre-scripts, 
> > execute them, preparing the applications for taking the
> > snapshot, and then 
> > it would flush the filesystem, kernel and disk caches as
> > needed to prepare for
> > a consistent snapshot.
> > 
> > Preparing the application for taking the snapshot might
> > mean something
> > like placing locks to database tables preventing disk
> > writes from happening, 
> > or shutting down the application, or whatever the specific
> > application requires.
> > 
> > post-snapshot action would again go through and execute all
> > the application specific
> > post-scripts releasing the applications back to normal
> > running state.
> > 
> > The application specific pre/post scripts could be stored
> > like this:
> > 
> >      
> > /etc/vss/scripts/xvda/001-mysql.pre
> >     /etc/vss/scripts/xvda/001-mysql.post
> > 
> >     /etc/vss/scripts/xvda/002-pgsql.pre
> >     /etc/vss/scripts/xvda/002-pgsql.post
> > 
> >    
> > /etc/vss/scripts/xvda/003-my_custom_app.pre
> >    
> > /etc/vss/scripts/xvda/003-my_custom_app.post
> > 
> > Or you could have "scripts/xvda2/*" if you have lvm volumes
> > directly as
> > partitions to the guest. You get the idea.
> > 
> > Then the next step would be to create a daemon ("vssd")
> > that listens for
> > requests from the network, and/or possibly from Xen dom0
> > through the hypervisor.
> > 
> > Or then the backup server might issue those pre-snapshot
> > and post-snapshot 
> > commands through ssh.. many options how to implement it.
> > 
> > Does this make any sense? Comments?
> > 
> > -- Pasi
> > 
> > > 
> > > > 
> > > > --- On Sun, 1/3/10, J Nb <j_nwb@xxxxxxxxx>
> > wrote:
> > > > 
> > > > > From: J Nb <j_nwb@xxxxxxxxx>
> > > > > Subject: coordinating VM snapshot with LVM
> > snapshot.
> > > > > To: xen-devel@xxxxxxxxxxxxxxxxxxx
> > > > > Date: Sunday, January 3, 2010, 6:17 PM
> > > > > Hi
> > > > >    How to co-ordinate VM snapshots and
> > disk
> > > > > snapshots ? Here is a sequence I am thinking
> > of
> > > > > automating...
> > > > > 
> > > > >   1.Tell VM to sync (say via ssh or some
> > special
> > > > > agent)
> > > > >      * This will do "sync" and any
> > > > > application specific hook to make it
> > quiescent.
> > > > > 
> > > > >   2. Tell VM/Hypervisor to flush all disk
> > buferes
> > > > > ?  (HOW ??? )
> > > > >      * Anything at this level?
> > > > > 
> > > > >   3. Pause the VM
> > > > > 
> > > > >   4. xm save -c (To save the VMs state in
> > to
> > > > > checkpoint file.)
> > > > >       (This may take more time !)
> > > > > 
> > > > >   5. Take LVM snapshot for each disk. 
> > > > > 
> > > > >   6. Unpause the VM
> > > > > 
> > > > >   7. Application level resume if any.
> > > > >   
> > > > >   7. Use the snapshot to copy each disks
> > ?!!!! (either
> > > > > dd or mount followed by cp/rsync)  (This
> > can be big!
> > > > > Any optimization suggestions )   
> > > > > 
> > > > >   8. discard snapshot.
> > > > > 
> > > > > 
> > > > > Would the above sequence work ? 
> > > > > 
> > > > > Any suggestions on individual steps ?
> > > > > 
> > > > > Anything specific for PV and HVM virtual
> > machines?
> > > > > 
> > > > > Is there a backdoor to qemu-dm console ?
> > > > > 
> > > > > Does anyone know how similar functionality
> > is implemented
> > > > > in XenServer product or Virtual Iron or
> > anyone else? 
> > > > > 
> > > > > -nb
> > > > > 
> > > > > 
> > > > >       
> > > > > 
> > > > 
> > 
> > 
> > _______________________________________________
> > Xen-users mailing list
> > Xen-users@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-users
> > 
> 
> 
> 
> 

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users