[Pacemaker] Fixed! - Re: Problem with dual-PDU fencing node with redundant PSUs

[Andrew Beekhof]

On 01/07/2013, at 10:06 PM, Vladislav Bogdanov <bubble at hoster-ok.com> wrote:

> 01.07.2013 14:53, Andrew Beekhof wrote:
>> 
>> On 01/07/2013, at 9:45 PM, Vladislav Bogdanov <bubble at hoster-ok.com> wrote:
>> 
>>> 01.07.2013 14:14, Andrew Beekhof wrote:
>>> ...
>>>>>> I'm yet to be convinced that having two PDUs is helping those people in the first place.
>>>>>> If it were actually useful, I suspect more than two/three people would have asked for it in the last decade.
>>>>> 
>>>>> I'm just silently waiting for this to happen.
>>>> 
>>>> Rarely a good plan.
>>> 
>>> ok, then here is my +1 :)
>>> 
>>>> Better to make my life so miserable that implementing it seems like a vacation in comparison :)
>>> 
>>> :)
>>> 
>>>> 
>>>>> Although I use different fencing scheme (and plan to use even more
>>>>> different one), that is very nice fall-back path for me. And I strongly
>>>>> prefer all complexities like reboot -> off-off-on-on to be hidden from
>>>>> the configuration. Naturally, that is task for the entity which has
>>>>> whole picture of what to do - stonithd. Just my 'IMHO'.
>>>> 
>>>> If the tides of public opinion change, then yes, stonithd is the place.
>>> 
>>> It would be natural.
>>> 
>>>> But I can't justify the effort for only a handful of deployments.
>>> 
>>> I do not use that only because I never used rgmanager, and that setup
>>> was not supported in pacemaker. If it was, I'd build my clusters in a
>>> different way, without need to reinvent a wheel. So, probably you may
>>> look from the other side - nobody uses unimplemented features but
>>> willing to use them once implemented.
>> 
>> Yes, but people around here also tend to be quite vocal when they think something is missing.
>> More so if its something critical.
> 
> ok, that is not critical (for me), there are always ways to work around.
> F.e. I plan (and I already did all hardware modifications, the only
> remaining part is an agent) to sit on reset lines (like rcd_serial does)
> with quido device from papouch (www.papouch.com) as a second-level
> fencing mech in addition to ipmi.
> 
> But, that would be nice to have feature if reboot command translation to
> multiple devices is implemented. And I would use it.

Well its possible right now, it "just" not super pretty to configure.
You should be able to leave out the location constraints though, that reduces the size a lot.

And if people start using it, then we might look at simplifying it.