[Pacemaker] live migration and colocation with lvm cloneset issue

Andrew Beekhof andrew at beekhof.net
Sun Sep 23 17:58:46 EDT 2012


On Thu, Sep 20, 2012 at 11:04 PM, denis bonnenfant
<denis.bonnenfant at diderot.org> wrote:
> Le 20/09/2012 00:02, denis bonnenfant a écrit :
>
> Hi,
>
> I have an issue with live migration of   a xen resource  over a clvm vg :
>
> - a lvm vg (cl_LVM1) cloned resource is defined on both nodes,
> - a xen resource (res_Xen_dns4) is using a lv from vg1, availiable on both
> nodes,
> - a colocation constraint is defined between domu and vg1
>
> When I try to live migrate res_Xen_dns4 from crm, it stops and restarts
> cloned instances of vg1, swapping them between nodes ! underlying drbd is
> demoted... Of course, this operation crashes all the other VMs connected to
> vg1, and finally all the cluster is fencing...
>
> Maybe I'm completely wrong, but I don't understand this behavior ! I guessed
> that the colocation constraints were defined with any clone member, and
> didn't force the clones to move. Did I missed something important ?
>
>
> OK, it seems that live migration implies no ordering/colocation constraints.

Not strictly true.
A resource can depend (colocation/ordering) on one or more clones and
still be able to migrate.
But as soon as something else depends on it, it can no longer be
migrated - because its not at the 'top' of the resource stack.

> I saw the serialize option in old threads, but found no info in the 1.1 doc.
>
>  So now how can I make my VMs to wait lvm to be available at startup, and to
> start on the node with active lvm (migration threshold ?).
> And in the reverse side, what will happen if a node goes to standby ? how to
> ensure that VMs will live-migrate *before* lvm/drbd will be stopped ?
>
> Denis
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
>




More information about the Pacemaker mailing list