[Pacemaker] Few questions

[Andrew Beekhof]

On Feb 9, 2009, at 12:25 PM, Romi Verma wrote:

>
>
> On Mon, Feb 9, 2009 at 4:04 PM, Andrew Beekhof <beekhof at gmail.com>  
> wrote:
> On Feb 9, 2009, at 11:21 AM, Dejan Muhamedagic wrote:
>
> On Mon, Feb 09, 2009 at 03:09:40PM +0530, Romi Verma wrote:
>
>
> sorry to interrupt in between. just one question from my side.  if
> no-quorum-policy is set to freeze then resource is not supposed to  
> stop .
>
> right.
>
> in
> that case node will not be fenced right??
>
> no-quorum-policy has nothing to do with fencing.
>
> well, depending on what it's set to, it can prevent fencing (no- 
> quorum-policy != ignore) or cause it to be initiated when quorum is  
> lost (no-quorum-policy == suicide)
>
> but if you're using freeze, and a resource must stop for other  
> reasons but fails to, then the node will not be shot (although it  
> probably should be).
>
> Thanks Andrew,
>
> in case of more then 2 nodes we cant use no-quorum-policy to   
> "ignore" as we cant just ignore quourum. as you said in case of  
> freeze node will not be fenced.

right, but thats a good thing.
It only applies if you have an even split (ie. 2-2) where no-one has  
quorum.
you only want to fence if you plan on taking over the other side's  
resources - which freeze wont do.

if you get a 3-1 split, then the side that has quorum _will_ shoot the  
1 node and take over its resources.

> so we have only two remaining suicide and stop. in case of suicide  
> node does not fence other  it just reset itself.

itself and anyone else in its current partition

> So the only remaining no-quorum-policy is "stop"  and if resource  
> fails to stop then the node should be fenced. am i right??

i think so
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20090209/2628fae9/attachment-0002.htm>