> > In RHEL 5 at least, anaconda requires the block device/file to be mapped
> > as disk (e.g hda, sda) not partition (e.g. sda1). Try changing it to
> > disk = [ "phy:lvm-raid/FileVolGroup,sda,w" ]
> Ok, so (at least in CentOS/RHEL?) the Xen domU sees the dom0 partition not
> as a partition but as a physical disk?
That depends on whether you put sda (whole disk) or sda1 (partition) in the
config quoted above. That's a standard Xen thing, not specific to
What I believe Fajar was saying is that the *installer* used by CentOS/RHEL
guests will be unhappy if you try and provide it with separate partitions
instead of a whole disk, which limits what you can do.
> sda1, sda, and hda all failed
> (sda1, sda were not visible while hda kept generating installer errors
> detecting the drives). hda1 shows up as weirdly named disk (/dev/hda1)
> that I need to partition out in the domU installer (ie, /dev/hda11,
> /dev/hda12, etc).
Wow. That's ... different to normal! I think we can safely say your
installer didn't like that, then!
> disk = [ "phy:lvm-raid/FileVolGroup,hda1,w",
> "phy:lvm-raid/io-swap,hda2,w" ] # assumption was that these
> were partitions...
They usually are, but it seems your guest's installer is expecting only a
whole virtual block device exported to it, not separate partitions. I think
that's probably resulting in the weird behaviour described above.
Another thing: assuming you're installing a PV guest, try using xvda instead
of hda. I'm not sure if this will solve your problem with the separate
partitions, but if you end up falling back to using a whole device it should
avoid the installer complaining so much!
> What concerns me is that if/when I want to grow disk available to a domU
> (this on in particular is a file server), I can't just grow what shows up
> as hda1, right? I'd need to add a new "disk" and extend a LVM within the
Alternatively you could extend the whole drive then resize the partitions
> Don't these layers of RAID+LVM (dom0) and Xen block device and LVM (domU)
> come at a price? I was hoping that the partition got mapped straight into
> the domU so I avoided any extra stuff within the domU and I would have the
> added bonus of being able to mount the drives and copy files between when
> setting this up.
Indeed. Xen supports this, your guest OS installer may not. It's probably
possible to "persuade" your guest OS to run on per-partition virtual devices
if you really want that but it would require a bit of extra fiddling.
I believe you can access partitions within an arbitary block device using the
kpartx tool, which may be useful for poking into a guest's virtual disk.
> > Having said that, if you're building lots of identical domUs, using
> > prebuilt template (and mappingthe block device as sda1 instead of sda)
> > should be faster than installer.
> If I can only do partitions, how does this work then? Unless all the
> drives are identical in size and I 'dd' the device??
I think you'd want to use a pre-installed template guest OS (possibly having
made modifications to make it run as you want) and then you'd copy it to
create new ones instead of doing an install+customise. You'd dd to duplicate
contents to other block devices in dom0, or cp to copy a file-based VBD.
Push Me Pull You - Distributed SCM tool (http://www.cl.cam.ac.uk/~maw48/pmpu/)
Xen-users mailing list