[Pacemaker] AP9606 fencing device

[Vadym Chepkov]

On Oct 27, 2010, at 8:11 AM, Dejan Muhamedagic wrote:

> Hi,
> 
> On Wed, Oct 27, 2010 at 01:58:20PM +0200, Pavlos Parissis wrote:
>> On 27 October 2010 13:43, Vadym Chepkov <vchepkov at gmail.com> wrote:
>> 
>>> 
>>> On Oct 27, 2010, at 7:27 AM, Pavlos Parissis wrote:
>>> 
>>>> On 27 October 2010 13:12, Vadym Chepkov <vchepkov at gmail.com> wrote:
>>>>> 
>>>>> On Oct 27, 2010, at 3:47 AM, Pavlos Parissis wrote:
>>>>>> 
>>>>>> Does anyone know any other PDU which works out of box with the
>>>>>> supplied stonith agents?
>>>>>> 
>>>>> 
>>>>> I use APC AP7901, works like a charm:
>>>>> 
>>>>> primitive pdu stonith:external/rackpdu \
>>>>>       params pduip="10.6.6.6" community="pdu-6" hostlist="AUTO"
>>>>> clone fencing pdu
>>>>> 
>>>>> Vadym
>>>> 
>>>> Then most likely the defaults OIDs of the rackpdu agents matches the
>>>> OIDs of the AP7901.
>>>> In my case I have to use OID for the device itself
>>>> 1.3.6.1.4.1.318.1.1.4.4.2.1.3  and OID for retrieving (snmpwalk) the
>>>> outlet list .1.3.6.1.4.1.318.1.1.4.4.2.1.4 .
>>>> 
>>>> Hold on a sec, are you using clone on AP7901? Does it support multiple
>>>> connections? Mine it doesn't.
>>> 
>>> Then it's useless regardless clone or not, you have to have multiple
>>> instances, because server can't reliable fence itself, right?
>>> 
>>> 
>>> 
>> My understanding is/was that I need to have one resource running on 1 of the
>> 3 nodes in the cluster and if a fence event has to be triggered then
>> pacemaker will send to it to the one stonith resource. I am planning to test
>> that the coming days.[1]
>> Am I right? if not then I have to buy a different PDU! :-(
> 
> Yes. In case a node which is currently running the stonith
> resource is to be fenced, then the stonith resource would move
> elsewhere first. But, yes, you should test this just like
> anything else. Make sure to test both the "node gone" event
> (failed links) and a critical action failing (such as stop).
> 
> Thanks,
> 
> Dejan

rackpdu stonith agent seems to explicitly remove node itself from list of hosts it can fence. so I assume if you have just one instance running, 
cluster would not see any stonith device capable to fence server where agent started initially. Would pacemaker move such resource anyway?
Since it reported it can't fence server in trouble?

Vadym