[CentOS] cluster - ip address lost when service stopped

Tue Dec 30 08:07:12 UTC 2008
Fabio Macchi <fabio.macchi at area.it>

Jay,

thanks for your answer, but there was a little misunderstanding:
the ip 10.0.181.4 is associated to the service and it's correct it goes away when service is stopped ( this is what I want, exactly like your environment ). My problem is that ip 10.0.181.41, that was the original ip associated to interface bound0 of the node AREA041, goes away too when service is stopped, and I don't understand why.
I suppose this is some issue related to fencing, as I don't use any fence device but I assign GFS resource internally to the service ( GFS share is mounted only when service start and only on the node hosting this service ): do you think this is a correct design ?

Tks
Fabio

-----Original Message-----
From: centos-bounces at centos.org [mailto:centos-bounces at centos.org] On Behalf Of Jay Leafey
Sent: lunedì 29 dicembre 2008 21:41
To: CentOS mailing list
Subject: Re: [CentOS] cluster - ip address lost when service stopped

Fabio Macchi wrote:
> Hi Gowrishankar,
> 
> this problem seems to be related to cluster, not to bonding: bonding is 
> working correctly, anyway I've tryied a test removing bonding, and I 
> experience the same problem directly on interface eth0.
> 
> This is my cluster.conf
> 
> <?xml version="1.0" ?>
> <cluster alias="cluster01" config_version="54" name="cluster01">
>         <fence_daemon clean_start="1" post_fail_delay="0" 
> post_join_delay="30"/>
>         <clusternodes>
>                 <clusternode name="AREA041" nodeid="2" votes="1">
>                         <fence/>
>                 </clusternode>
>                 <clusternode name="AREA042" nodeid="3" votes="1">
>                         <fence/>
>                 </clusternode>
>         </clusternodes>
>         <cman expected_votes="1" two_node="1"/>
>         <fencedevices/>
>         <rm>
>                 <failoverdomains>
>                         <failoverdomain name="httpd failover domain" 
> ordered="0" restricted="1">
>                                 <failoverdomainnode name="AREA041" 
> priority="1"/>
>                         </failoverdomain>
>                 </failoverdomains>
>                 <resources>
>                         <ip address="10.0.181.3" monitor_link="1"/>
>                 </resources>
>                 <service autostart="0" domain="httpd failover domain" 
> name="Apache" recovery="disable">
>                         <script file="/etc/rc.d/init.d/httpd" 
> name="script httpd"/>
>                         <ip ref="10.0.181.3"/>
>                 </service>
>                 <service autostart="0" domain="httpd failover domain" 
> name="Service Mail" recovery="disable">
>                         <script file="/etc/rc.d/init.d/MailScanner" 
> name="MailScanner"/>
>                         <clusterfs device="/dev/DATI_MAIL/DATI_MAIL" 
> force_unmount="1" fsid="5845" fstype="gfs2" mountpoint="/dati_mail" 
> name="Share_dati_mail" options=""/>
>                         <ip address="10.0.181.4" monitor_link="1"/>
>                 </service>
>         </rm>
> </cluster>
> 
> Many thanks
> 

 From what I can tell, the behaviour you are noticing is consistent with 
your cluster.conf file.  Since you have made the IP addresses part of 
the service definitions, the IP would "go away" when the associated 
service is stopped.  If the service moved to another node, however, the 
IP would be enabled on the host to which the service was moved.

If you want the IP addresses to be independent of the service state, 
then add them using files in /etc/sysconfig/network-scripts to define 
the "alias" addresses and remove them from your service definitions. 
See /usr/share/doc/initscripts-*/sysconfig.txt for details on how to set 
up the alias addresses.

I'm doing something similar with IP addresses in a cluster, but I WANT 
the IP address to migrate to the target host when a service is moved 
from one node in the cluster to another.  I have the IP address 
resources tied to the individual services to make that happen.

Hope that helps!
-- 
Jay Leafey - Memphis, TN
jay.leafey at mindless.com