[Linux-cluster] fencing device

[Jakub Suchy]

Hi,

> |fencing can be a SPOF, you need two fencing devices).
> 
> ILO is used on many sites successfully.

It is. Just just have to be careful when using it.

> |- It's not a shared device. It MUST be on the same network path as
> |  heartbeat.
> Not true...  I have my ILO on my public network and my heartbeat on a private network switch.  No problems with fencing in this configuration.

Talking about two node clusters: Try disconnecting your private network.
Each node tries to fence each other, iLO is not shared and is on public
network, therefore it may succeed. If it accidentally happens in the
same time (and it can), all nodes will fence each other and you end up
with powered off servers. This is what you don't want to happen...

> I have some ILO issues to be fixed in RHEL-5.3 for fencing on blades...   The situation I have seen is you must be successful in power off and power on for the ILO fence to complete as you suggest.  I hope this will would be changed and that a ILO power off would be sufficient.  In the case of "away for a planned repair" you would not perform an ILO fence, rather you would shutdown the node in the cluster gui.

Yes. And now the node is offline and out for a planned repair. How do
you reboot the active node now? If you do: it starts, waits few seconds
for the offline node to join and then fence it = problem is here.

Cheers,
Jakub

-- 
Jakub Suchý <jakub.suchy at enlogit.cz>
GSM: +420 - 777 817 949

Enlogit s.r.o, U Cukrovaru 509/4, 400 07 Ústí nad Labem
tel.: +420 - 474 745 159, fax: +420 - 474 745 160
e-mail: info at enlogit.cz, web: http://www.enlogit.cz
Red Hat Premium Business Partner