[Pacemaker] migrate/failover and md device?

[Miles Fidelman]

I'm running a 2-node failover cluster that has a fairly traditional 
configuration - using DRBD to maintain synchronize data across 2 nodes, 
and pacemaker to failover Xen virtual machines. Underneath it all, I'm  
running software RAID10 on the individual nodes.

I've been thinking of adding a couple more servers to my rack, and
looking for ways to set up a more general migration/failover scheme -
and I keep coming back to DRBD as a limitation: even with a cluster file
system, without a SAN, it looks like synchronizing 2 nodes is my limit.

Which leads me to wonder about spreading a software RAID volume (md)
across drives published via AoE or iSCSI, configured to survive a node
failure that takes multiple drives offline.

Which leads to several questions about how to assemble and manage the
resulting RAID device. Assuming a separate gigE or 10gigE network
linking all the nodes, onto which all nodes publish their drives:

1. The simple case would be to assemble and mount the RAID device on the
same node as the VM, then use pacemaker to manage the md device using
Pacemaker. There seem to be several resource agents for managing RAID
devices, but each one seems to be either commented with lots of caveats,
and/or deprecated. So... has anybody done this? Does it work? How's
the performance?

2. An alternate model would be to do something with a cluster file
system - to keep all the components mounted, on all nodes in the
cluster, with failover just happening automatically if a node and its
drives die. But I'm sort of at a loss on how to proceed here.
Suggestions? Pointers to learning resources?

Thanks much,

Miles Fidelman



-- 
In theory, there is no difference between theory and practice.
In<fnord>   practice, there is.   .... Yogi Berra