[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] Adding a new fence agent



Hi,
Yes you can issue a "poweroff server 3" to gracefully shutdown a blade
(I'm using "poweroff server # force",  to ensure a fast shutdown,
without the "force" option sometimes the OS shutdown hanged and things
didn't work properly.

I'm not sure how it works internally, I'll have to shutdown or disable
the iLO port manually to test it since it is internal to the blade
enclosure.
I'll be able to do more tests and clean up the script next week.

I modified the script to ensure I have another fence level besides the
iLO port.
During failover tests we removed one of the blades and the cluster went
"kaput" because it couldn't access the iLO port of the blade and fence it.
With another fence level, if it fails to reach the iLO port, it will try
to fence the node through the enclosure and the cluster recovers normally.

In this perspective it works as a redundancy if the iLO port fails or
there is no communication with the blade.



Marcos David


Gianluca Cecchi wrote:
> I have a c7000 too with two test blades I'm going to install.
> I'm available to test it if you like.
> My planned OS will be RedHat EL 5 U3 x86_64 with its clustersuite
> Blades will be 2 x  BL685c G1 serving Oracle 10gR2
> At this moment the fw version  of the c7000 is 2.25, while iLo fw is 1.60
>
> One question: I see that on standard blade fence agent (that is for
> IBM enclosures) you can issue a command such as "power off blade
> number 3".
> I imagine the same on this c7000 fence agent, correct?
> In this case can we say that commands on c7000 (onboard admin I
> presume) bypasses the iLO of the blades or not? So are they
> effectively an alternative to iLO commands if for example the iLo is
> broken, or not?
>
> One other thing. Probably you reported the cluster.conf xml layout
> only as an example, because at this moment the c7000 fence agent would
> try to do its work only if the iLO agent fails, so you have to invert
> their order inside the <clusternode> sections
>
> bye,
> Gianluca
>
> --
> Linux-cluster mailing list
> Linux-cluster redhat com
> https://www.redhat.com/mailman/listinfo/linux-cluster
>
>
>   



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]