[Linux-cluster] Network failure results clusterenvironmentunstable & fragile
Pena, Francisco Javier
francisco_javier.pena at roche.com
Tue Feb 28 14:03:17 UTC 2006
> On Mon, 2006-02-27 at 20:28 -0500, Lon Hohberger wrote:
> > On Mon, 2006-02-27 at 09:22 +0100, Pena, Francisco Javier wrote:
> > > Hi Deval,
> > >
> > > If you are using iLO fencing, you could try the latest
> fence package
> > > (1.32.10). I have seen a similar problem, and it is
> because recent
> > > iLO firmware versions behave a little different (they try
> to make a
> > > soft restart instead of a hard reboot).
> > Also, if you're using iLO fencing, make sure you boot with acpi=off
> > (see /etc/grub.conf) and have acpid stopped. (e.g.
> chkconfig --level
> > 2345 acpid off)
> > ACPI soft-poweroff (which iLO tries to do) is exactly what
> you do not
> > want in a cluster.
> Wow, this sounds fairly important, but I do not see this
> mentioned in the RHCS-4 "Configuring and Managing a Cluster"
> documentation when I look at the setup for ILO fencing.
> Where is this documented so others deploying HP Proliants
> don't fall into a trap? :)
Well, as far as I have been able to check, this is not necessary if you
running the latest version of the fence package. Previous versions with
iLO firmware (1.75+) caused this ACPI soft-poweroff issue (they were
a power button press, instead of a "press and hold for a few seconds").
I would just be happy if they added the perl-Crypt-SSLeay package as a
Dependency to the fence package. If it is not installed, you may have a
surprise when trying to fence off a node using the iLO :).
More information about the Linux-cluster