[Linux-cluster] Fencing Race Question

[Lon Hohberger]

On Fri, 2007-10-26 at 08:47 -0700, Scott Becker wrote:
> To clarify, example #2 in the faq: redundant power feeds, redundant APCs 
> - no single point of failure. Provides High availability but complicates 
> fencing and race prevention.
> 
> This necessitates two power switch logins to successfully fence. I'll 
> test by manually logging in to one device and tell luci to fence a node. 
> With my manual login getting in the way, the fence operation should fail 
> and turn off no ports and not prevent manual login to other apc, no 
> matter which switch I manually log into. I haven't got a chance to test 
> this yet and I wanted to tap into your wisdom to test my logic.

If you log in to switch #2, the cluster will be able to turn off ports
on switch #1, but the fencing operation will fail.

-- Lon