Ipfail configuration

[Kevin Dwyer]

On Wed, 12 Feb 2003 15:31:13 -0000
Tim Saunders <Tim.Saunders@aquilauk.co.uk> wrote:

> Adding "respawn hacluster /usr/lib/heartbeat/ipfail" to ha.cf worked
> but I also had to add an hacluster user, put that user in the haclient
> group and chown /var/lib/heartbeat/api/ipfail.req and
> /var/lib/heartbeat/api/ipfail.req to hacluster as they had uid 17 for
> which my systems have no user.

Hm, I thought heartbeat was doing that on make install these days. 
Perhaps only in the CVS version.

> I also had to put "nice_failback on" into ha.cf. This surprised me as
> I don't really see why its nessisary (except for legacy design/code
> reasons). If Ipfail is in use and has detected a failed interface I
> would expect it to mark that node as failed and thus the node would
> not be eligible for being the active node (unless the other node had
> the same interface failed). Thus I would expect the other node to take
> over.

Ah sorry about that.  I assumed you already had it set, but yes it is
necessary.  The reason is that we want to give ipfail the controls over
whether to failover.  Without the nice_failback behavior, hearbeat would
immediately failback the resources because for all it's concerned, the
node is still alive (we can communicate.)  ipfail alters this
assumption, and induces failovers because of network failures.

> Thanks for your help. Ipfail is now working perfectly.

No problem, glad it works.

-- 
/* kevin@pheared.net               http://pheared.net/devel/ */
/* Network Security Engineer       http://pheared.net/~kevin */
/* Sabotage will set us free.   Throw a rock in the machine. */
/*   >++++++++++[<++++++++++>-]<.+++++.----.[-]++++++++++.   */