[Linux-HA] resource stickyness

[aichains]

Hello,

I have a simple 2-node cluster, where i want to setup a floating IP on
RHEL4u5-32bit.  I am using the heartbeat RPMs found in
http://dev.centos.org/centos/4/testing/i386/RPMS/

Here are the contents of the relevant config files.

[root at master ha.d]# cat ha.cf
bcast                   eth0
auto_failback off
keepalive               2
deadtime                10
node                    master.example.com
node                    slave.example.com
respawn                 hacluster /usr/lib/heartbeat/ipfail
apiauth                 ipfail gid=haclient uid=hacluster
traditional_compression false
use_logd                yes

[root at master ha.d]# cat haresources
master.example.com 10.0.0.252

[root at master ha.d]# cat /etc/hosts
127.0.0.1               localhost.localdomain localhost
10.0.0.250              master master.example.com 
10.0.0.251              slave slave.example.com 
10.0.0.252		vip vip.example.com

When I bring heartbeat up on the master, it assumes the IP resource.
When I bring up heartbeat on the slave, it sees that the master has the
IP and "just sits there".
When I run hb_standby on the master, the resource correctly transfers to
the slave.
When I run hb_takeover on the slave, the resource correctly transfers to
the slave.

Now the "problem"...

When I run "ifdown eth0" on the master to simulate a dead node, the
resource correctly transfers to the slave.  However, when I run "ifup
eth0" on the master, the resource transfers back to the master.  I
configured ha.cf with "auto_failback off" under the assumption that this
would allow the resource to stay on the slave node even after the master
comes back online.  I referenced http://www.linux-ha.org/ActivePassive
for that assumption.

Can anyone tell me how to make the IP stay on the "slave node" even
after the master comes back online ?



-- 
aichains <aichains at nonstophate.com>