[Pacemaker] new doc about stonith/fencing

[Dejan Muhamedagic]

Hi,

On Mon, May 04, 2009 at 11:37:42AM +0200, Peter Kruse wrote:
> Hi Dejan,
>
> Dejan Muhamedagic wrote:
>> As usual, constructive criticism/suggestions/etc are welcome.
>
> Thanks for sharing.
> Allow me to bring up a topic that to my point of view is important.
> You have written:
>
>> The lights-out devices (IBM RSA, HP iLO, Dell DRAC) are becoming increasingly popular
>> and in future they may even become standard equipment of of-the-shelf computers.
>> They are, however, inferior to UPS devices, because they share a power supply with their
>> host (a cluster node). If a node stays without power, the device supposed to control it
>> would be just as useless. Even though this is obvious to us, the cluster manager is not
>> in the know and will try to fence the node in vain. This will continue forever because all
>> other resource operations would wait for the fencing/stonith operation to succeed.
>
> This is the same problem with PDUs as they share the same power supply with
> the host as well.

Never worked with PDUs, but there should be a difference compared
to the lights-out devices: if a PDU looses power then all nodes
connected to it do as well. I suppose that that is much more
likely than only one outlet failing. In that case all cluster is
down and you have other things to worry about rather than fencing.

> Is there any intention to deal with this issue?  I'm
> thinking of the powerfail algorithm:
>
> If the PDUs becomes unavailable and shortly after the host is unavailable as
> well, then assume the host is down and fenced successfully.
>
> This would be true if the PDU (and with it the host) loses power.
> At the moment it looks that stonith without such an algorithm is
> a SPoF by design, because after a single failure (powerloss), the
> cluster is not able to bring up the resources again.

This is something certainly worth looking at. On the other side,
it's not really so difficult or pricey to devise a setup where
you won't have to worry about this detail.

Cheers,

Dejan

> Looking forward to your comments,
>
>   Peter
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker