[Linux-cluster] Re: CS5 two-nodes with quorum disk

[Alain Moulle]

Hi Lon

Many thanks for your last both responses, it's much clearer
in my head now ... ;-)

When you speak of "a method to enable master-set-wins (e.g. for
operation w/o heuristics)" , will it be delivered by a patch when
ready ? or just a matter in configuration file ?
if patch, when do you think it could be available ? because
I'm interested ... I guess it will be sort of a way to establish
that one of the two-nodes cluster is master and in case of
"perfect symmetrical splits", it is always the master the winner ...
if so, that will be fine for me .

Thanks again
Alain


From: Lon Hohberger <lhh at redhat.com>
Subject: Re: [Linux-cluster] Re: CS5 two-nodes with quorum disk
To: linux clustering <linux-cluster at redhat.com>
Message-ID: <1197915584.4959.22.camel at ayanami.boston.devel.redhat.com>
Content-Type: text/plain

On Mon, 2007-12-17 at 10:36 +0100, Alain Moulle wrote:

>> Hi Lon
>>
>> I've carefully read your last detailed information. I've a
>> better understanding but something is again not clear for me :
>> in my two node cluster node1/node2, with quorum disk , without any heuristic,


Need a heuristic for now.


>> I would like to be sure that if there is a failure on the heart-beat
>> network, only one node fences the other and not both, so :
>> when I do on node2 ifdown on eth if of heart-beat, what is the
>> mechanism via the quorum disk that assures that ?
>> Or how must I configure to assure that ?


What does your "heartbeat network" look like?

You need a way to tell the difference between "I am alive" and "I am
alive but should be dead".

In your case, a simple link-detection heuristic would be an easy way to
do this - no router / pinging required: no link = "I cannot be the
master node" => immediately loses the race.

If the heartbeat link isn't available, the node should remove itself.

Obviously, this will not work with a crossover cable (both nodes would
think they're dead if the link was unplugged).  Switches are cheap.

Now there are lots of other ways to determine who should win.  However,
when thinking about this, it is very important that understand that
perfect symmetrical splits - i.e. splits where there is absolutely no
discernible difference between the nodes - will usually end in a race.

Currently, I'm working on a method to enable master-set-wins (e.g. for
operation w/o heuristics); I'll post when it's complete.

-- Lon