[Pacemaker] Drbd dstate to trigger failover

[Lars Ellenberg]

On Wed, Apr 20, 2011 at 09:49:13PM -0400, Shravan Mishra wrote:
> Wherever there is estates/estate it should be dstates/dstate.
> 
> Thanks
> Shravan
> 
> On Wed, Apr 20, 2011 at 9:46 PM, Shravan Mishra
> <shravan.mishra at gmail.com> wrote:
> > Hi,
> >
> > I'm using following config for io errors:
> >
> >
> >
> > resource resource {
> >  disk {
> >    on-io-error detach;
> >    ...
> >  }
> >  ...
> > }
> >
> >
> > The above leads to following  state in case of disk errors:
> >
> > Diskless/UpToDate
> >
> > Under drbd documentation there is a following line:
> >
> > STMT -- "If the disk failure has occured on your primary node, you may
> > combine this step with a switch-over operation."
> >
> > When I look at drbd resource agent's monitor:
> >
> > drbd_monitor() {
> >        local status
> >
> >        drbd_status
> >        status=$?
> >
> >        drbd_update_master_score
> >
> >        return $status
> > }
> >
> >
> > Now in the above function the error is reported only based on role not
> > on estate. drbd_update_master_score is only updating score based on
> > dstates.

drbd_update_master_score absolutely takes the role into account as well.
It triggers on role, local disk state and remote disk state.

A Diskless Primary connected to an UpToDate peer
will receive a master score of 10,
whereas the UpToDate peer will have a master score of 10000.

Depending on other constraints and settings, that may or may not be
enough to trigger a switchover right there.

> > My question is can I return error in drbd_monitor if my primary goes
> > Diskless and then cause the switchover based on the STMT?
> >
> > Or  should I be doing something else?
> >
> >
> > Sincerely
> > Shravan
> >

-- 
: 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.