[Linux-cluster] Fencing: Prevent rebooting halted node

[Digimer]

On 03/21/2012 07:23 AM, emmanuel segura wrote:
> Hello Nicolas
> 
> The first i can recommend it's use a quorum disk and for your problem i
> know that it's called fencing-loop

Quorum disk works only when someone has a SAN proper, and even then, I
am pretty sure that a node in an unknown state will still need to be fenced.

> you got two choices
> 
> 1:don't put the daemons cluster at boot time

This delays the issue, not avoid it. When Nicolas starts cman, it will
still fence the peer.

> 2:If you use a qdisk you can use clean_start="1" in the fence_daemon
> tag, this can be used if you are not using gfs or gfs2 or
> <cmantwo_node="1"expected_votes="1"/>

I feel *very* uncomfortable with the 'clean_start="1"' option. Making
any kind of assumption about the state of a node in a cluster is asking
for trouble.

-- 
Alteeve's Niche!
Madison Kelly        647-501-5200
Papers and Projects: https://alteeve.com