[Linux-cluster] How to run same service in parallel in RedHat Cluster 5.0

[Ruben Sajnovetzky]

We crossed e-mails :)
I sent the new-fresh configuration.
I thought about fencing, the problem is that we have a very "odd"
configuration because I don't really need fence anything ... Maybe I
can establish kind of rule like "If Central is not working, Collector cant
work" or similar, will think on it.

The IP will not be migrated, we have a requirement to have Virtual IP for
the Central (only), that is why it is configured.

Regarding FS (other member asked why GFS), I don't see the advantage
because, again, we are not "sharing" the storage.

Thanks



On 28-Sep-2011 12:50 PM, "Digimer" <linux at alteeve.com> wrote:

> On 09/28/2011 06:20 AM, Ruben Sajnovetzky wrote:
>> <?xml version="1.0"?>
>> <cluster alias="PPM_CL1" config_version="81" name="PPM_CL1">
>>         <fence_daemon clean_start="0" post_fail_delay="0"
>> post_join_delay="30"/>
>>         <clusternodes>
>>                 <clusternode name="server-87111" nodeid="1" votes="2">
>>                         <fence/>
>>                 </clusternode>
>>                 <clusternode name="server-87112" nodeid="2" votes="1">
>>                         <fence/>
>>                 </clusternode>
>>         </clusternodes>
>>         <cman expected_votes="1">
>>                 <multicast addr="224.4.5.6"/>
>>         </cman>
>>         <fencedevices/>
>>         <rm>
>>                 <failoverdomains>
>>                         <failoverdomain name="PPM_GW_FDN" nofailback="1"
>> ordered="0" restricted="1">
>>                                 <failoverdomainnode name="server-87111"
>> priority="1"/>
>>                         </failoverdomain>
>>                         <failoverdomain name="PPM_UNIT_FDN" nofailback="1"
>> ordered="0" restricted="1">
>>                                 <failoverdomainnode name="server-87112"
>> priority="1"/>
>>                         </failoverdomain>
>>                 </failoverdomains>
>>                 <resources>
>>                         <ip address="10.116.159.105" monitor_link="1"/>
>>                         <script file="/usr/local/bin/cluster/ppm_gw_ha"
>> name="PPM_GW"/>
>>                         <fs device="/dev/mapper/VolGroup00-optvol"
>> force_fsck="1" force_unmount="0" fsid="36845" fstype="ext3"
>> mountpoint="/opt" name="PPM_OPT_FS" self_fence="0"/>
>>                         <apache config_file="conf/httpd.conf" name="web"
>> server_root="/etc/httpd" shutdown_wait="0"/>
>>                         <clusterfs device="/dev/hdd" force_unmount="0"
>> fsid="20023" fstype="gfs" mountpoint="/mnt" name="pru1" self_fence="0"/>
>>                 </resources>
>>                 <service autostart="0" exclusive="0" name="PPM_PRUEBA"/>
>>                 <service autostart="0" domain="PPM_UNIT_FDN" exclusive="0"
>> name="PPM Units">
>>                         <fs ref="PPM_OPT_FS"/>
>>                 </service>
>>                 <service autostart="0" domain="PPM_GW_FDN" exclusive="0"
>> name="PPM Gateway">
>>                         <fs ref="PPM_OPT_FS">
>>                                 <ip ref="10.116.159.105">
>>                                         <script ref="PPM_GW"/>
>>                                 </ip>
>>                         </fs>
>>                 </service>
>>         </rm>
>> </cluster>
> 
> Ah, here it is.
> 
> If I can recommend, remove the parts you're not yet using (the empty
> service, the apache resource, etc). Remove any options that aren't
> critical... make it as simple as possible. Once you get it working, you
> can start adding things back.
> 
> As an aside, with the config you have, the IP address will never
> migrate. It will come up on and only ever run on PPM_GW_FDN. Also, it
> will not come up if the fs resource fails as it is a child. This may be
> what you want though.
> 
> Also, you *must* configure fencing. Even without a shared file system, a
> fence call will hang the cluster. It can cause many odd symptoms. Get
> fencing configured and working, strip down the config and try again. If
> it still fails, repost the config as it is at that point.