[Linux-cluster] Pseudo-Fencing for 2-node DRBD+GFS

[Gordan Bobic]

Hi,

Thanks for the replies on the other thread. I'm still assimilating them. 
Meanwhile, however, I am faced with a different problem. It turns out 
that the servers I've got to prototype this with only have MegaRAC 780 
(a.k.a. DRAC 2) management cards in them. The problem with this is that, 
to the best of my ability to tell, these don't seem to provide a sanely 
usable interface to allow for the machine to be powered down via the 
command line (Windows only management utility - I can't even get it to 
ping, only arping, and nmap isn't coming up with anything!)

So, I need a poor man's pseudo-fencing hack that will protect the shared 
file system by stopping the DRBD replication between the hosts (range of 
options available including using iptables to firewall the cluster NIC, 
or just downing the interface completely, as it's not used for anything 
else). The theoretical assumption (I don't really have a choice...) is 
that if one of the machines fails, it will fail completely and not 
attempt to fight over the resources that need to be failed over (in this 
case just IP addresses).

Is there a minimum API that a fencing script has to implement to be 
usable via cluster.conf? If so, can anyone please point me at some 
documentation for it? Would this be a reasonable way to implement 
fencing in a 2-node case?

Thanks.

Gordan