[Linux-cluster] Two-node cluster: Node attempts stateful merge after clean reboot

[David Teigland]

On Wed, Sep 11, 2013 at 01:03:07PM +0200, Pascal Ehlert wrote:
> Now when I switch node 2 back on, it looks like it joins the cluster and
> is kicked out immediately again, which again results in fencing of node
> 2. I have purposely set the post_join_delay to a high value, but it
> didn't help.

You may want to look at increasing FENCED_MEMBER_DELAY in
/etc/sysconfig/cman.  The value you choose should be greater than the
maximum time corosync takes to discover another node after starting.

Corosync creates a fully functional 1-node cluster upon starting and
before discovering other nodes.  This behavior can cause problems,
particularly in two node clusters, which the setting above works around.
It would be consistent with the problem you're seeing.

Dave