[Linux-HA] Redundant Rings "Still Not There?"

[Vadym Chepkov]

On Oct 23, 2010, at 4:12 PM, Robinson, Eric wrote:

>>> "3-node cluster" is much easier to say than to configure,
>>> apparently. :-)
> 
>> It really isn't :) 
> 
> Encouraged by your "it really isn't," I now press forward.  :-)
> 
> Based on what I'm hearing, this is what I think I have learned...
> 
> It is possible to build a 3-node cluster with redundant heartbeat paths,
> but I need to use two switches, something like this:
> 
> 
> -------------SWITCHED NET #1------(192.168.10.0/24)-----------------
> 	   |                   |                           |
>         |                   |                           |
> |-----------------| |-----------------|       |---------------------|
> |        |        | |        |        |       |          |          |
> |   |--bond0--|   | |   |--bond0--|   |       |     |--bond0--|     |
> | eth0       eth1 | | eth0       eth1 |       |   eth0       eth1   |
> |                 | |                 |       |                     |
> |     SERVER_A    | |     SERVER_B    |       |       SERVER_C      |
> |                 | |                 |       |                     |
> |      eth2       | |       eth2      |       |        eth2         |
> |        |        | |        |        |       |          |          |
> |-----------------| |-----------------|       |---------------------|
>         |                   |                           |
>         |                   |                           |
> -------------SWITCHED NET #2-------(192.168.20.0/24)-----------------
> 
> 
> That way, if something happens to switched network #1, Corosync can
> still track node status through switched net #2.
> 
> Once this configuration is built, I can use Pacemaker with resource
> constraints to ensure that resource R1 can only run on SERVER_A or
> SERVER_C (usually A) and resource R2 can only run on SERVER_B and
> SERVER_C (usually C) and SERVER_C acts as a failover for both resources.
> 
> Is this correct? 
> 

That's correct. And since it's a 3 node cluster you can make this using simple constraints like this:

location R1-prefers-A R1 100: SERVER_A
location R1-not-B -inf: SERVER_B
location R2-prefers-B R2 100: SERVER_B
location R2-not-A -inf: SERVER_A

Vadym