On Wed, Oct 24, 2007 at 12:09:04PM +0200, Pierre-Alain RIVIERE wrote:
> I've also added some udev custom rules to have the same device name on
> each box
This isn't actually necessary. The same name makes things easier from the user/admin
perspective, but the LVM side of things uses the ID of the pv, rather than the device name.
> This is a normal behavior? And when I will deploy CMAN, CLVM and iSCSI
> on another box what will be LVM behaviours? I can't guarantee that
> /dev/sdj will be accessible on the other box, only the symlink will be
> the same.
LVM is using /etc/lvm/lvm.conf (on this RH machine anyway) to choose what devices to scan.
The default on this RHEL5 machine is:
devices {
dir = "/dev"
scan = [ "/dev" ]
filter = [ "a/.*/" ]
cache = "/etc/lvm/.cache"
write_cache_state = 1
sysfs_scan = 1
md_component_detection = 1
}
which means that it'll detect anything in /dev and use that. This has bit me in the past
(using drbd) but shouldn't be a problem here. If it is (and you're not using LVM for the
OS) you could change this to only scan your links instead.
The reason it caused a problem with DRBD (it's a way of doing mirroring between 2 nodes, if
you've never seen it) is that the PV is visible as the local device as well as the mirrored
one, and writing changes to the local copy doesn't work (they don't get replicated to the
other node).
So as long as the /dev/sd is a valid device to use, don't worry about it.
Indeed - if your iSCSI stuff is as straight forward as it sounds, I'd be tempted to stop
doing the symlinks and just let the LVM stuff do its "thing".
Graham
--
Linux-cluster mailing list
Linux-cluster@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/linux-cluster
