Hi all,

I’m trying to setup a cluster of 2 machines with Centos 5.2 to host a postfix+spamassassin+clamav+mailscanner service. Below cluster software versions

rgmanager.i386                           2.0.38-2.el5_2.1       installed

cman.i386                                2.0.84-2.el5_2.2       installed

Every machine (hp blade server ) has 4 interfaces, bounded in this way:

Eth0, eth1 -> bond0 -> connection for public service ( 10.0.181.x )

Eth2,eth3  -> bond1 -> connection for intra-cluster communication  ( 192.168.44.x )

bond0     Link encap:Ethernet  HWaddr 00:21:5A:48:DA:BE 

          inet addr:10.0.181.41  Bcast:10.0.181.255  Mask:255.255.255.0

          inet6 addr: fe80::221:5aff:fe48:dabe/64 Scope:Link

          UP BROADCAST RUNNING MASTER MULTICAST  MTU:1500  Metric:1

          RX packets:85 errors:0 dropped:0 overruns:0 frame:0

          TX packets:86 errors:0 dropped:0 overruns:0 carrier:0

          collisions:0 txqueuelen:0

          RX bytes:12963 (12.6 KiB)  TX bytes:9144 (8.9 KiB)

bond1     Link encap:Ethernet  HWaddr 00:1F:29:6D:7D:08 

          inet addr:192.168.44.41  Bcast:192.168.44.255  Mask:255.255.255.0

          inet6 addr: fe80::21f:29ff:fe6d:7d08/64 Scope:Link

          UP BROADCAST RUNNING MASTER MULTICAST  MTU:1500  Metric:1

          RX packets:29 errors:0 dropped:0 overruns:0 frame:0

          TX packets:223 errors:0 dropped:0 overruns:0 carrier:0

          collisions:0 txqueuelen:0

          RX bytes:4612 (4.5 KiB)  TX bytes:31746 (31.0 KiB)

Then I’ve created a new Mail service with these local resources:

- Ip address 10.0.181.3

- Script /etc/rc.d/init.d/MailScanner

- GFS file system on a SAN

Service start, but the problem is that, when I stop the service, external ip address is removed from bond0.

What may cause this ? I don’t find any helpful information in /var/log/messages, what may I check to investigate ?

I’ve not setted any fence device, as I have configured GFS resource internally to the service ( GFS partition is mounted only from the server running the service ): is this a wrong design ?

Has anyone already setup this kind of service on a Centos cluster ?

Many thanks in advance for any hints.

Fabio