[Linux-cluster] Avoiding fencing loops with startup quorum (patch)

[David Teigland]

On Fri, Oct 17, 2008 at 06:11:53PM +0200, Federico Simoncelli wrote:
> On Fri, Oct 17, 2008 at 5:41 PM, David Teigland <teigland at redhat.com> wrote:
> > Here's another possibility I hadn't thought of before:
> >
> > . don't set two_node/expteced_votes in cluster.conf
> > . edit init.d/cman and possibly /etc/sysconfig/cman to do
> >  cman_tool join -w        (joins cluster and waits to be a member)
> >  cman_tool wait -q        (waits for quorum, both nodes to be members)
> >  cman_tool expected -e 1  (change expected votes to 1)
> 
> I tried this before but the downsides were:
> 
> - long waits due not  being in quorum (fence timeout is 600 seconds by
> default)

cman_tool join -w can will possibly wait a long time if the other node
is not up or is partitioned.  That's the price you pay for avoiding the
potential back-and-forth fencing loop.

> - you have to rewrite the cluster.conf to make it work

I don't know what you're refering to.  You simply don't include the
two_node/expected_votes line.

> - break legacies with distros/systems not using this init script

Yes, you have to hack the init script.  There's a change coming in RHEL5.3
that's very similar to this, where you won't need to hack anything:

http://git.fedorahosted.org/git/cluster.git?p=cluster.git;a=commitdiff;h=5ea416d26ec2b6bf605c573a5173736d0f8cd27c

http://git.fedorahosted.org/git/cluster.git?p=cluster.git;a=commitdiff;h=397b8111d2d69b9dd25e7b074822be571f274032