[Linux-cluster] Split-brain with DRBD active-active + RHCS

[Digimer]

On 03/15/2011 01:58 PM, Bazy wrote:
> Hi Mark,
> 
> Yes, clustered file system is mandatory. Even with gfs(2) DRBD will
> not recover by itself from a split brain. I think specific options are
> needed in drbd.conf "after-sb-0pri after-sb-1pri after-sb-2pri", but
> don't know what the exact ones are.
> 
> Best regards!

For the recovery;

resource r0 {
	device /dev/drbd0;

	net {
		after-sb-0pri discard-zero-changes;
		after-sb-1pri discard-secondary;
		after-sb-2pri disconnect;
	}
}

To tie it into fenced via cman (to fence instead of split-brain), also add:

resource r0 {
	device /dev/drbd0;
	
	disk {
		fencing resource-and-stonith;
	}

	handlers {
		outdate-peer "/sbin/obliterate";
	}
}

You can download 'obliterate' from here:

http://people.redhat.com/lhh/obliterate
(found here: http://gfs.wikidev.net/DRBD_Cookbook)

See also:

http://www.drbd.org/users-guide/ch-rhcs.html

-- 
Digimer
E-Mail: digimer at alteeve.com
AN!Whitepapers: http://alteeve.com
Node Assassin:  http://nodeassassin.org