[DRBD-user] Adapting a new fence device for DRBD

Florian Haas florian.haas at linbit.com
Wed Dec 1 17:38:05 CET 2010

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


Hi,

sorry this one does indeed seem to have got lost in the shuffle.

On 2010-11-05 19:22, Digimer wrote:
> Hi all,
> 
>   I've created a home-brew, open source fence device based on Red Hat's
> FenceAgentAPI. The fence device and agent are now well supported by Red
> Hat and now I would like to extend the support to DRBD.
> 
>   I've come up with a semi-canned two-node cluster using DRBD and Red
> Hat's cluster suite 3. The one thread left dangling is that, when DRBD
> detects a split brain, there is no fence action taken or recovery process.
> 
>   I understand that DRBD has the ability to tie into a fence device, and
> I assume that doing so would help prevent split-brain conditions as one
> node can fence the other, mark itself as "more" UpTpDate and then
> invalidate the peer if and when it recovers from the fence action. Is
> this a correct understanding?
> 
>   Any further information would be much appreciated. :)

Take a look at the drbd.conf man page for the fence-peer handler and the
"fencing resource-and-stonith" option in "disk". Essentially what
happens is this:

- DRBD detects peer is gone
- DRBD freezes I/O
- DRBD fires the fence-peer handler and observes its exit code as per
the convention explained in
http://www.drbd.org/users-guide/s-fence-peer.html
- if all is well (peer is definitely no longer accessing the disk), DRBD
un-freezes I/O and resumes normal operations.

For Pacemaker, we can have the fence-peer handler tie in directly with
the stonith-ng fencing daemon, and the cluster infrastructure does the
rest for us. So once NA plays nicely with the Pacemaker STONITH API, you
get DRBD fencing support almost for free.

All of this is obviously specifically needed in dual-Primary
configurations. For single-Primary DRBD it is not _strictly_ necessary
as we tend to do quite well with the Outdated status for a Secondary
that has lost its Primary peer. The DRBD Pacemaker OCF resource agent
has this covered.

Does this help?

Cheers,
Florian

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20101201/35569f63/attachment.pgp>


More information about the drbd-user mailing list