[Linux-cluster] rgmanager on debian

[Lon Hohberger]

On Wed, Jul 11, 2007 at 10:41:41AM +0400, Pavel D. Kuzin wrote:
> This problem was fixed.

That's good.

> Seems it was because in fstab was record /dev/Hosting/Hosting_data and 
> rgmanager tryes to mount /dev/mapper/Hosting-Hosting_data.

That's strange; fstab entries shouldn't matter.

> Now i have another question.
> How i can start one service on both nodes simultaneously ?

There's no way to start two copies of the same service... but:

* clusterfs can be referenced multiple times, and
* scripts can be referenced multiple times.

(Note: IP addresses can not be referenced multiple times and/or started
on two nodes...)

So, you can build something like:

  <resources>
    <clusterfs name="mygfs" device... mountpoint... options...
      force_unmount="0"/> <!-- force_unmount=0 is important -->
	<script name="myscript" path.../>
  <service name="foo-instance1">
    <clusterfs ref="mygfs"/>
    <script ref="myscript"/>
    <ip address="10.1.1.1"/>
  </service>
  <service name="foo-instance2">
    <clusterfs ref="mygfs"/>
    <script ref="myscript"/>
    <ip address="10.1.1.2"/>
  </service>

The only thing to note is that often times, the application "myscript"
might try to bind to INADDR_ANY - preventing the two services from
running on the same node.  Additionally, most scripts will stomp on each
other during the "stop" phase if stopped on the same node - something to
be aware of.  (You can get around this by storing say
OCF_RESKEY_service_name and using that to control instances individually
in your script...)

Generally, the two services should be able to operate independently -
even if sharing resources

-- 
Lon Hohberger - Software Engineer - Red Hat, Inc.