Hi,

On Thu, Feb 21, 2008 at 09:00:35PM +0100, Johan Hoeke wrote:
> Dejan Muhamedagic wrote:
> > Hi,
> > 
> > On Thu, Feb 21, 2008 at 04:09:19PM +0100, Johan Hoeke wrote:
> >> Dejan Muhamedagic wrote:
> >>> Hi,
> >>>
> >>> On Thu, Feb 21, 2008 at 01:26:12PM +0100, Johan Hoeke wrote:
> >>>> Dejan Muhamedagic wrote:
> >>>>> On Thu, Feb 21, 2008 at 12:19:55AM +0100, Johan Hoeke wrote:
> >>>>> Oops. So there's an on_fail=fence for this monitor operation. Is
> >>>>> that necessary?
> >>>> We want the cluster to failover if oracle breaks for whatever reason.
> >>>> At least I think we do ;)
> >>> But failing over is not the same as fencing. Why would you fence
> >>> a cooperating node.
> >> Hi Dejan,
> >>
> >> We're using fibre channel attached storage. I want to fence to protect
> >> the data on the SAN.
> > 
> > That's a good cause, but why do you think that the data is
> > jeopardized. I mean, if the node is healthy and if it says that
> > it doesn't have the disks mounted I suppose that you could
> > trust it. That's why fencing is by default done only in case a
> > resource can't be stopped. Of course, it is another matter if you
> > think that the resource agents may lie or that something else in
> > the setup can't be trusted.
> > 
> > Thanks,
> > 
> > Dejan
> > 
> 
> OK, I understand. I'll change from monitor on_fail=fence to stop
> on_fail=fence and test,test,test.

on_fail=fence is default for stop operations as those failures
are dangerous.

> I have to be super careful that the
> SAN filesystem doesn't get corrupted again. That happened the other day
> by accident when a wrong ipfilter config was pushed by mistake. The
> heartbeat interface was filtered out, a split brain situation occurred
> and the SAN filesystem was corrupted. Stonith didn't save us for
> whatever reason.

You have to have a reliable stonith device. Do you think that
on_fail=fence in the monitor op would have made the situation
better?

> The application managers don't have much confidence in
> heartbeat since then. :(

That's a shame.

Thanks,

Dejan

> regards,
> 
> Johan
> 



> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to