On 24 November 2010 09:48, Xavier Montagutelli <xavier.montagutelli@xxxxxxxxx> wrote: > On Wednesday 24 November 2010 09:34:48 Jankowski, Chris wrote: >> Xavier, >> >> Thank you for the explanation. >> This all makes sense. >> >> One more question about one of the documents you pointed me to: >> >> What does this do exactly and why do I need it: >> >> Quote: >> >> 4) Update your initrd on all your cluster machines. Example: >> prompt> new-kernel-pkg --mkinitrd \ >> --initrdfile=/boot/initrd-halvm-`uname -r`.img --install `uname -r` >> >> Unquote > > Caution : the following are only supposition, because I haven't read the > lvm.sh script. > > In step 3 of http://sources.redhat.com/cluster/wiki/LVMFailover , you have to > modify your lvm.conf file, "volume_list" parameter, to filter the VG/LV that can > be activated on a particular host. > > In step 4, they say to create a new initrd : I suppose this step is necesary > to include the modified lvm.conf file inside the initrd, and to NOT activate the > VG located on the shared storage at boot time. > > The lvm.sh script must add or remove the "good" tag (i.e. a tag matching the > hostname of the node running the service) on the fly. > > If someone can confirm or give additional pointers ? That's how I understand it. I've used LVM on RHEL5 *without* clvmd and not had any problems with corruption, etc., but I haven't used snapshots. You have to try really, really hard to break the tagged LVM config from the command line, as the tags prevent activation (which also prevents you from accidentally mounting the FS on different nodes). It's worth knowing about the "--config" argument to the LVM commands, and how to active the LVs from the command line so you can do maintenance to the tagged VG/LVs outside of RHCS: $ lvchange --config "activation { volume_list = [ '@$HOSTNAME' ] }" vg00/test -a y I am not a LVM hacker, so take the following comments with the appropriate caution: If you only ever make changes to the LVM on the active node, I think you'd have to be really unlucky to suffer corruption due to stale LVM metadata. Although if you're carrying out long running tasks like relocating PEs, it might be worth freezing the service (I think this is still probably overly cautious). I think if you start making changes on multiple nodes at the same time, you will suffer badly (but the tags should stop this from happening accidentally). If you are activating LV resources on the basis of their VG, LVM snapshots should survive the resource being relocated between nodes; when the VG is deactivated on the original node, both the original and snapshots LV will be deactivated at the same time, so you won't miss any writes in the snapshot. In my RHEL6 test environment, I just created a two node cluster with cman/clvmd and could create a snapshot on the LVs in a shared VG. This fails under RHEL5. I'm not sure I'd trust it to actually work though... It's probably worth directing some of your questions at the LVM list for a more definitive answer: https://www.redhat.com/mailman/listinfo/linux-lvm PS: I'm sure you will, but you should test it :) Cheers >> >> Regards, >> >> Chris Jankowski > -- > Xavier Montagutelli Tel : +33 (0)5 55 45 77 20 > Service Commun Informatique Fax : +33 (0)5 55 45 75 95 > Universite de Limoges > 123, avenue Albert Thomas > 87060 Limoges cedex -- Jonathan Barber <jonathan.barber@xxxxxxxxx> -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
