From mboxrd@z Thu Jan  1 00:00:00 1970
From: Lon Hohberger <lhh@redhat.com>
Date: Mon, 18 Feb 2013 14:33:13 -0500
Subject: [Cluster-devel] qdisk and fence race
In-Reply-To: <24E144B8C0207547AD09C467A8259F7557A510DA@lisa.maurer-it.com>
References: <24E144B8C0207547AD09C467A8259F7557A510DA@lisa.maurer-it.com>
Message-ID: <512281F9.9080901@redhat.com>
List-Id: <cluster-devel.redhat.com>
To: cluster-devel.redhat.com
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit

On 02/13/2013 12:32 PM, Dietmar Maurer wrote:
> The following example is copied from 'man qdisk':
> 
>        <cman two_node="0" expected_votes="3" .../>
>         <clusternodes>
>             <clusternode name="node1" votes="1" ... />
>             <clusternode name="node2" votes="1" ... />
>         </clusternodes>
>         <quorumd interval="1" tko="10" votes="1" label="testing">
>             <heuristic   program="ping  A  -c1  -w1"  score="1"  interval="2" tko="3"/>
>         </quorumd>
> 
> Is it true that this can still result is a fence race?
> 
> If so, what is the suggested way to avoid that? Should I configure with  master_wins="1" instead?

It's best to use master_wins over a heuristic.  Ping as a tiebreaker
seems good in practice, but the problem is that there are certain
network failures where two nodes can see the thing they're pinging - but
not each other (e.g. some switch loops).

master_wins protects against this.

-- Lon