[Linux-cluster] Failover IPs and Interfaces

[gordan at bobich.net]

On Tue, 11 Mar 2008, gordan at bobich.net wrote:

> How can I specify in cluster.conf which physical interface the failover 
> domain IPs should be assigned to on each node? I cannot seem to find any 
> documentation on the subject. If anyone can point me in the right direction, 
> that would be most appreciated.

Perhaps it wasn't immediately obvios what I meant here, so I'll expand 
with an example:

In cluster.conf:
[...]
<rm>
   <failoverdomains>
     <failoverdomain name = "private-ip" ordered = "1" restricted = "1">
       <failoverdomainnode name = "host1" priority = "1"/>
       <failoverdomainnode name = "host2" priority = "2"/>
     </failoverdomain>
     <resources>
       <ip address = "192.168.255.250" monitor_link="1"/>
     </resources>
     <service autostart = "1" domain = "private-ip" name = "myservice">
       <ip ref = "192.168.255.250"/>
     </service>
</rm>
[...]

What I cannot seem to find documented is how to control which physical 
interface the ip resource 192.168.255.250 is going to be assigned to on 
fail-over. How can this be defined, and/or inferred, if there are multiple 
interfaces in the system? Is there an option that allows something like 
this:

<ip address = "192.168.255.250" monitor_link="1" interface="eth3" />

or

<ip ref = "192.168.255.250" interface="eth3" />

The option instead of "interface" is what I am after if there is such a 
thing. Is there such a thing? If not, how can resource-to-interface 
binding be defined?

TIA.

Gordan