[DRBD-user] 2 x 2-node configuration?

[Jaap Winius]

Quoting Jaap Winius <jwinius at umrk.nl>:

> My RHEL7 project is considering using DRBD to create sets of 2 x  
> 2-nodes that replicate synchronously across two data centers. The  
> first half of each set would consist of a primary and a secondary  
> with the latter taking over in case of failure (Heartbeat-based  
> Pacemaker cluster). However, the same block device(s) should also  
> replicate to the second half of the set in the second data center.  
> Crucially, it should be possible to turn these roles around without  
> too much effort when the choice is made to make the second data  
> center active and the first one passive.
>
> Is this possible with DRBD and if so, what kind of replication  
> configuration should be used?

Yes. First, install DRBD9 on all four nodes to create a single  
replicating DRBD cluster. Next, install the pcs and pacemaker packages  
(which include Corosync as dependency instead of Heartbeat) on the  
same four nodes, but don't use them to create a single  
Corosync/Pacemaker cluster. Rather, create two clusters -- one in each  
data center -- of two nodes each, ensuring that DRBD continues to  
replicate across all four nodes and that only one of the two  
Corosync/Pacemaker clusters is active at a time. This way, failovers  
and failbacks will only involve two nodes in the data center where the  
Corosync/Pacemaker cluster is active. Use your preferred configuration  
management software to stop the Corosync/Pacemaker cluster in one data  
center and start it in the other, automatically moving the DRBD  
primary along with it.

Cheers,

Jaap