[Linux-cluster] 2-node tie-breaking

[gordan at bobich.net]

On Fri, 8 Feb 2008, Lon Hohberger wrote:

>>>>> Well, here's a cheap 'out' ;)
>>>>> Ebay item # 250213910258
>>>>> 8 port WTI NPS for $70 + $15 s/h.
>>>>
>>>> Indeed, that would be tempting if it wasn't on the wrong side of the
>>>> atlantic. A more local search for similar things doesn't seem to come up
>>>> with anything. :-(
>>
>> *Don't flame me, i didn't want to spam the list with commerial stuff but
>> thought this might help you out*
>>
>> If your interested in a powerswitch solution i would recommend you to look
>> at the solutions from www.lubenco.nl
>>
>> They have 8 port (C13) 10 amp powerswitches with load monitoring and
>> snmp configuration support for about 117 euro a pice. (new) If your
>> willing to buy in bulk the price even drops well below 100 euro.
>
> Guys, this is good stuff - the problem is we need to make an agent that
> can talk to them before we can really solve Gordan's problem.  Does
> anyone want to volunteer?   I'll write up a fence agent API "howto"
> today ...

Thanks for all this, guys, much appreciated. :-)

I just had another thought. Most motherboards these days are ATX, which 
means ATX type "short-pins-to-power-on" power switches. That means that as 
a _REALLY_ cheap solution I could just get something like a small relay 
switch and wire it into the serial port. When a pin on RS232 goes high 
(e.g. DTR), it activates the switch. I think it would be pretty reliable, 
and the total cost of components would be pennies. The fence agent would 
also be a total of about 10 lines of code, too. :)

Can anyone think of why this wouldn't work? OK, it involves a bit of DIY, 
but it would work for a 2-node cluster. With more nodes, pulsing the power 
switch repeatedly might bring the node back up, which would be less than 
ideal if it's _really_ nastily intermittent. But since this isn't really 
scalable to cluster with more than 3 nodes (2 serial ports per node), it 
probably doesn't matter.

Gordan