Here’s what I’ve found:
If I vgchange the 2 affected VGs to inactive, multipath will pick up the devices. However, once I do that, I can’t seem to get the VGs reactivated in such a way to make the volumes mountable. I use vgchange –a y to make the VG active, then do “vgscan —mknodes”. That doesn’t seem to create the devices in /dev, only in /dev/mapper, so I’ve tried issuing /etc/init.d/udev-post reload, which does create the volumes in /dev/<vg>/<lv>. The issue is, it has no blkid. And if I try to mount the mapper device directly, mount doesn’t know what filesystem type it is. I’m not an lvm pro, so it’s possible I’m missing a step or two, but in the past when moving around snapshots of lvms, vgscan —mknodes has always been sufficient.
As for not having multipath in initrd, I have created an initrd image with the following:
mkinitrd /boot/initramfs-with-multipath.img 2.6.33.8-149.fc13.x86_64 --with=dm-multipath
I then created a grub entry for it, booted off it, and no change in behavior. Is that what you had in mind?
-Brian
On 8/27/10 11:26 AM, "Malahal Naineni" <malahal@xxxxxxxxxx> wrote:
If you can, deactivate the affected VG, restart multipath (it should be
able to load maps now) and then activate the VG. If this works, then
there is something that is causing LVM to claim devices before multipath
does.
Thanks, Malahal.
PS: If you don't have multipath in initrd, LVM *may* claim paths as PV's
before the multipath from active root can claim them. Don't have enough
info on Fedora13 initrd if this can happen though.
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
-- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
