Отвечу сам себе. Может кому пригодится.

crm_resource -C -V --force -r xen_r5 -H dom0b

решило ситуацию.

А это объяснение:

>>> "Taldevkar, Chetan" <chetan.taldevkar at patni.com> 18.07.2007 07:23 >>>
>Hi all,
>
>When I start cluster lunixha is able to invoke start call on both the
>nodes. On first node start fails as script calls echo "stopped"
>followed by exit 1 as this resource needs to be running on the second
>node. After that it successfully starts on second node. But if I
>simulate error conditions on second node it does not invoke the script
>on first node resulting no failover.
>

As far as I know: As soon as a RA couldn't be started on a node this
RA is not able to start there anymore. You have to use crm_resource -C
(--force) to reset the information that this RA couldn't be started there.
So, if you like the resource to be run initially on node 2 add a
location constraint to the resource and it will be started on the
second node. After a failure of that resource heartbeat will start
it on the first one if and only if the values for stickiness and
failure stickiness are set properly.

Best regards
Andreas Mock