On 06/09/2011 08:33 AM, Jun'ichi Nomura wrote:
Hi
On 06/08/11 17:01, Osier Yang wrote:
This is to address BZ# https://bugzilla.redhat.com/show_bug.cgi?id=702260, though even if with this patch, the user might see error like "Unable to deactivate logical volume",
Can you try the attached patch? (in addition to the upstream lvm fix you mentioned below)
Recent distributions generate udev events *after* the use of devices through "watch" rules. As a result, lvremove/lvchange can race with it and fail to remove/deactivate volume.
I haven't tested the patch but I could fix the similar problem of lvremove by putting 'udevadm settle' before that. So I think it's worth trying.
it could fix the problem if the lv is referred to by another existing LVs, allowing the user remove the lv successfully without seeing error like "Can't remove open logical volume".
For the error "Unable to deactivate logical volume", libvirt can't do more, it's problem of lvm, see BZ#: https://bugzilla.redhat.com/show_bug.cgi?id=570359
And the patch applied to upstream lvm to fix it: https://www.redhat.com/archives/lvm-devel/2011-May/msg00025.html
This lvm patch fixes only the case where lvremove itself generates udev events. You'll still see the problem as the udev events will be generated when a VM finishes using the volume, for example.
I'm worried about using "udevadm settle" will introduce some delays, but it looks to me this is the only workable fix for the bug, as for "lvchange -an $vol; lvremove $vol", it could encount the similiar problem with "lvremove -f", as you can still get the udev processing event when doing "lvchange -an". Anybody has thoughts on this? if no, I'd think Jun'ichi's patch is nice, and push it. Thanks Osier