[libvirt-users] fsck and guest images

[Daniel P. Berrange]

On Wed, Jul 03, 2013 at 12:24:24PM +0200, Nicolas Sebrecht wrote:
> The 03/07/13, Jamie Fargen wrote:
> 
> > Hey!
> > 
> > I have some RHEL6 hypervisors and the VMs are in raw qemu image files in a
> > local raid array linux raid + lvm + ext3. When a kernel update is installed
> > a reboot is necessary, usually it has been more than 180 days since the
> > last reboot and the file system is fsck'd and this takes 2-3 hours.
> > 
> > I am curious to know if there is any documentation that addresses the pro's
> > and con's of fsck'ing the volume containing /var/lib/libvirt/images.
> 
> This is standard ext fsck to prevent errors after some time.
> 
> In order to avoid long fsck times, we use ext4 almost everywhere (for
> both hypervisors and guests). I would suggest you to switch to a newer
> filesystem supporting fast fsck.
> 
> > Could fsck make a change to the underlying file system that the guest
> > images are stored on, which the guest operating system may not be able to
> > handle when it runs its own file system maintenance, i.e. fsck or chkdsk.
> 
> Talking about filesystems errors, you should assume that everything is
> possible. Though, the fsck on /var/lib/libvirt/images is limited to the
> filesystem used by the hypervisor and should not interfere with the
> filesystems of the guests. An exception I'm aware of is a reiserfs
> filesystem inside another reiserfs filesystem (/var/lib/libvirt/images
> is reiserfs and the guests are in reiserfs, too).
> 
> > Is file system maintenance on the hypervisor volume storing the VM images
> > redundant to the VM's own file system consistancy utilities.
> 
> As said above, it's not redondant. The fsck at hypervisor level keeps
> limited to the filesystem at hypervisor level.

Of course if you think that having 2 fscks is overkill, then you could
also change the way you storage images on the host. eg, use an LVM volume
for each guest disk, instead of storing them in /var/lib/libvirt/images.
That way the guest OS is 100% responsible for its data integrity and you
don't waste I/O bandwidth by having duplicate fscks in host & guest.


Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|