[Pacemaker] AP9606 fencing device

[Pavlos Parissis]

On 28 October 2010 10:21, Dejan Muhamedagic <dejanmm at fastmail.fm> wrote:

> Hi,
>
> On Wed, Oct 27, 2010 at 08:15:09PM +0200, Pavlos Parissis wrote:
> > On 27 October 2010 19:46, Pavlos Parissis <pavlos.parissis at gmail.com>
> wrote:
> >
> > > I did more testing using the clone type of fencing and worked as I
> > > expected.
> > >
> > > test1 hack init script to return 1 on stop and run a crm resource move
> on
> > > that resource
> > > result node it was fenced and resource was started on the other node
> > >
> > > test2 using firewall to break the heartbeat links on node with resource
> > > result node it was fenced and resource was started on the other node
> > >
> > > As Dejan suggested I am going to run the same type of tests when 1
> fence
> > > resource is used.
> > > In this test I will try to cause a fencing on the node which has
> fencing
> > > resource running on it and see if pacemaker moves the resource before
> it
> > > fences the node.
> > >
> > >
> > >
> > >
> > I did the same tests without cloning and pacemaker moves fencing resource
> > before triggers a reboot on the node where fencing resource was running.
> > So, cloning fencing resource and having just one fence resource have the
> > same behaviour! at least for these 2 tests.
> > now I don't know which configuration solution I should choose!
>
> Whichever you feel more comfortable with, providing that the
> device really can support multiple connections simultaneously.
> I'd opt for non-cloned version. It's simpler, it avoids possible
> device contention.
>
> Thanks,
>
>
Under which conditions does pacemaker initiate multiple connections to a
fencing device?
Given the fact that rackpdu agent uses SNMP, so connections limits
<shouldn't> apply, I don't quite understand how cloned version will give me
issues. I make a big assumption here that connections limits are not
applicable when fencing device is contacted over SNMP  .

Furthermore, using cloned version a fence event triggers faster than
non-cloned version, because in non-cloned version situation the resource
must move to another node, if the node to be fenced holds the fencing
resources.

Because of the above, I selected for now the cloned version. But your mail
worries a bit.

what test can I do in order to make sure that the cloned version will not
give me issues?

Cheers,
Pavlos
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20101028/b27a4eb0/attachment-0002.htm>