[Linux-cluster] Conga issue

Ryan McCabe rmccabe at redhat.com
Tue Nov 13 20:01:57 UTC 2007


On Tue, Nov 13, 2007 at 11:47:17AM -0800, Scott Becker wrote:
> I'm still not sure 'why' the pair is needed, it's an implementation 
> detail of fencing. Do some fencing devices have different commands for 
> turning back on? Or does it just allow the ability to configure "auto 
> fence off, but only manual fence on" by deleting the on items. If this 
> is the case then it seems that there should be a check box in luci to 
> that effect.

When a node is to be fenced, the <device> blocks within the <method>
tags are processed in order. If no 'option' attribute is specified, the
default action is to issue a power off, then a power on. When a node has
redundant power supplies, it's necessary to power them all off before
powering any back on, or the machine will never lose power, defeating
the point of trying to fencing it. So if you're using power fencing and
you have N power supplies for a node, you'll end up with N*2 <device>
blocks in the fence method block for that node; the first N/2 having
attribute 'option' set to off, and the last N/2 having the attribute
'option' set to on.


Ryan




More information about the Linux-cluster mailing list