florian.haas at linbit
Dec 1, 2009, 4:58 AM
Post #2 of 2
(1094 views)
Permalink
|
|
Re: Xen, live migration and fencing with pacemaker
[In reply to]
|
|
On 2009-11-30 13:41, infernix wrote:
> Thats all fine and dandy and it seems to work, until fencing comes into
> play. I've tried to use the crm-fence-peer.sh and unfence-peer.sh
> scripts, but they fail to execute succesfully because they look at the
> CIB and try to find the master for a given drbd resource. Since I don't
> have to add these to CRM, but the (un)fence scripts do require them to
> work correctly, I get errors like:
>
> crm-fence-peer.sh[5804]: invoked for r0
> cibadmin: [5808]: info: Invoked: cibadmin -Ql
> crm-fence-peer.sh[5804]: WARNING drbd-fencing could not determine the
> master id of drbd resource r0
> kernel: block drbd0: helper command: /sbin/drbdadm fence-peer minor-0
> exit code 1 (0x100)
> kernel: block drbd0: fence-peer helper broken, returned 1
> kernel: block drbd0: State change failed: Refusing to be Primary while
> peer is not outdated
>
> This obviously isn't what I'd like to happen :)
Obviously. :) The crm-fence-peer.sh script requires that there is a
master/slave resource whose constraints it can manipulate. If you are
using the "drbd:resource" disk syntax in your Xen config, then that
master/slave resource is not required, and there are no matching
constraints that the crm-fence-peer.sh handler can work with.
> So the question is, what's the right way to do fencing in a xen
> live-migrating cluster?
On Heartbeat based Pacemaker clusters, you can use dopd. See
http://www.drbd.org/users-guide/s-heartbeat-dopd.html.
And, since during live migration you have a short time period where DRBD
is Primary on both nodes, you can add an extra layer of protection by
enabling STONITH in Pacemaker.
> And could the Xen chapter get updated with this
> info?
Yes, this is going on my to-do list.
Cheers,
Florian
|
signature.asc
