[Pacemaker] Strange behaviour of dual master DRBD

[Lars Ellenberg]

On Mon, Sep 14, 2009 at 10:23:55AM +0200, Michael Schwartzkopff wrote:
> Am Montag, 14. September 2009 09:51:45 schrieb Lars Ellenberg:
> > On Fri, Sep 11, 2009 at 09:50:56AM +0200, Michael Schwartzkopff wrote:
> > > Am Donnerstag, 10. September 2009 19:25:53 schrieb Lars Ellenberg:
> > > > On Thu, Sep 10, 2009 at 05:10:39PM +0200, Michael Schwartzkopff wrote:
> > > > > Hi,
> > > > >
> > > > > I configured a dual master DRBD-8.3.2. When the nodes stat there is
> > > > > no problem. Both DRBD become master. But when I set on node to
> > > > > standby and wake it up again, the DRBD on that node stays secondary
> > > > > and does not become master.
> > > > >
> > > > > My config:
> > > > > primitive resDRBD ocf:linbit:drbd params drbd_resource="r0"
> > > > > ms msDRBD resDRBD meta notify="true" master-max="2"
> > > > >
> > > > > No further constraints.
> > > > >
> > > > > When the second node is online again ptest -sL shows:
> > > > > (...)
> > > > > resDRBD:0 promotion score on suse2: 50
> > > > > resDRBD:1 promotion scpre on suse1: -1
> > > > >
> > > > > Since the "-1" prevents the resource from beein promoted, I
> > > > > understand the behaviour of the cluster, but why isn't the resource
> > > > > beeing allowed th become master on that node?
> > > > >
> > > > > Thanks for any enlightenting answers.
> > > >
> > > > Most likely it prevents you from shoting yourself in the foot ;)
> > > >
> > > > look at /proc/drbd and the kernel logs (appart from the ha.log, of
> > > > course) on the drbd nodes to find out more.
> > > > I bet you manoevered yourself into diverging data sets (aka DRBD "split
> > > > brain").
> > > >
> > > > If it turns out to be a drbd.ocf bug,
> > > > let me know.
> > >
> > > Hi,
> > >
> > > another confusing fact:
> > >
> > > The primitive DRBD was created without a monitor operation. When I add a
> > > monitor operation, better two (Master,Slave) the described problem goes
> > > away. But in my understanding the complete setup should work as expected
> > > WITHOUT the monitoring operation.
> >
> > Well, your understanding is wrong.
> > If you want the details,
> > there is a lot of magic going on associated with the master scores,
> > see the drbd.ocf script, drbd_update_master_score.
> >
> > You should definitely configure notify.
> > You may get away without notify, if you configure monitor,
> > and wait for a few monitor actions to happen.
> 
> Hi,
> 
> I configured the notification. My resource look like:
> primitive resDRBD ocf:linbit:drbd \
> 	params drbd_resource="r0"
> ms msDRBD resDRBD \
> 	meta resource-stickiness="100" notify="true" master-max="2" clone-max="2" 
> interleave="true"
> 
> After putting one one to standby and online again I still get:
> resDRBD:0 promotion score on suse1: 50
> resDRBD:1 promotion score on suse2: -1
> 
> So the resource will NOT get promoted. Only with monitoring this behaviour 
> disappears.

Fine.
So configure monitor.

-- 
: 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.