[Pacemaker] Unnecessarily Failover when restarting network!!!

[Lars Ellenberg]

On Thu, Mar 04, 2010 at 10:54:57AM +0100, Andrew Beekhof wrote:
> On Mon, Mar 1, 2010 at 5:41 PM, Jayakrishnan <jayakrishnanlll at gmail.com> wrote:
> >
> >
> > On Mon, Mar 1, 2010 at 4:04 PM, Andrew Beekhof <andrew at beekhof.net> wrote:
> >>
> >> On Thu, Feb 25, 2010 at 7:32 PM, Jayakrishnan <jayakrishnanlll at gmail.com>
> >> wrote:
> >> > Hiiii,
> >> > One more question...
> >> > I managed make every things to work with Heartbeat-Pacemaker (2.99 -
> >> > 1.0.5).
> >> > I have a cluster ip, pingd, postgresql lsb and a lsb resource
> >> > successfully
> >> > configured for slony replication .. But when I restart network via
> >> >
> >> > # /etc/init.d/networking restart
> >> >
> >> > split-brain is happening.. I have increases my monitor intervels and
> >> > even
> >> > dampening in all resources and even in by ha.cf file but still split
> >> > brain
> >> > is happening. Please advice me!!!
> >>
> >> I guess its permanently affecting Heartbeat's communication or
> >> membership mechanisms.
> >
> > Thanks for your reply Andrew....
> > Yea it should be ,,, My heartbeat link is via a cross over cables directly
> > connected... How about increasing the 'keepalive' time or 'deadtime'
> > again....
> >
> > IN my ha.cf.deadtime' is 60sec, warntime is 50sec...
> > In my pingd resource i have given dampen as 60sec, monitor interval as
> > 65sec, ie, greater than dead time in ha.cf... ohhh.... did i missed it... My
> > dead time in ha.cf is same as dampen in pingd... Could it be the issue...
> > However it is not supposed to give a failover when I restart my network,,
> > naaa?
> 
> To be honest, I've not used heartbeat seriously for a few years now.
> So I don't really feel qualified to talk about its ha.cf settings much.
> 
> pingd dampen settings can't affect heartbeat's ability to communicate
> with its peers though.

If you do "ifconfig down" or equivalent, like you do for network restart,
all heartbeat network related communication channels are broken, and
will not recover without manually restarting them.

"Don't do that, then."


-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.