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

Lon Hohberger lhh at redhat.com
Tue Dec 18 23:22:26 UTC 2007


On Tue, 2007-12-18 at 14:00 +0100, Alain Moulle wrote:

> Me again, always fighting with my two node cluster with quorum
disk ;-)
> I've not yet received your response to my last email if you have already
> answered, because I 'm in digest mode, so , sorry if  ...

https://www.redhat.com/archives/linux-cluster/2007-December/msg00188.html


> Meaning that in cluster.conf I should have :
>      <quorumd interval="1" label="QDISK_5_0" min_score="1" tko="10" votes="1">
>          <heuristic interval="10" program="ping -t1 -c1 <another host>" score="1"/>
>      </quorumd>
> 
> Am I right when telling that this heuristic is mandatory to avoid
> race-to-fence and eventually dual-fencing ?
> I hope so ... it would be the end for me about quorum disk ... ;-)

Correct, you need a heuristic for now.

In your case, you could also consider a heuristic which checks the
ethernet link too (using ethtool for example - rather than the typical
'ping') if there is no other host on the network you want to monitor.

Also - your heuristic should be more like one of the following:

  <heuristic interval="2"
             tko="3" 
             program="ping -t1 -c1 <another host>"
             score="1"/>

  <heuristic interval="2"
             program="ping -t3 -c1 <another host>"
             score="1"/>

Reason: You don't want a single ICMP packet to determine node fitness.
If that ping gets lost (network being full, or any reason really), the
node will commit suicide.  (The man page probably needs updating about
that!)

-- Lon





More information about the Linux-cluster mailing list