Hi,

On Tue, Feb 26, 2008 at 08:04:58PM +0900, Keisuke MORI wrote:
> Hi,
> 
> Dejan Muhamedagic <[EMAIL PROTECTED]> writes:
> 
> > Hi,
> >
> > On Tue, Feb 26, 2008 at 11:52:16AM +0900, Keisuke MORI wrote:
> >> Hi,
> >> 
> >> Johan Hoeke <[EMAIL PROTECTED]> writes:
> >> > Dejan Muhamedagic wrote:
> >> >> Hi,
> >> >> 
> >> >> On Fri, Feb 22, 2008 at 05:59:23PM +0100, Johan Hoeke wrote:
> >> >>> LS,
> >> >>>
> >> >>> Anybody here using some kind of brocade fencing with heartbeat like
> >> >>> RedHat offers in its cluster software? I found this reference:
> >> >>> http://linux.die.net/man/8/fence_brocade
> >> >>>
> >> >>> It turns out to be the attached perl script. Got it from
> >> >>> http://mirror.centos.org/centos/4/csgfs/i386/RPMS/fence-1.32.50-2.el4.centos.1.i686.rpm
> >> >>>
> >> >>> Would it be possible to use this as an external stonith script?
> >> >> 
> >> >> No, because stonith is about fencing nodes and this would be
> >> >> fencing resources.
> >> >
> >> > Point taken!
> >> 
> >> Fencing nodes by isolating I/O is very interesting idea though.
> >
> > It's a bit of a terminology problem, but it boils to the same
> > thing. Fencing a resource means building a fence around it as
> > your sf-ex does. Fencing a node is done by bringing it down so
> > that it can't connect to any resource anymore. Stonith is used
> > only on the node level.
> 
> I believe that fence_brocade is the node level fencing, but am I wrong?

It could be, but it is more like resource level. Depends on the
number of f/o switch ports used.

> In my understanding, fence_brocade does fencing by bringing all
> the Fibre connections of the node down to protect the data instead
> of turning the power off.
>
> It's a different approach from the SF-EX to avoid the data corruption
> but I think it's an interesting alternative.

Well, actually it could work, but it would be an exception in
terms of behaviour. I guess that one could use the poweroff
command to mean detach a SAN port. Looking at the man page of
fence_brocade it would be easy to implement a stonith external
plugin to use it.

> >> >  > I think that right now the only way would be to implement an RA
> >> >> which would fence resource. That's what Junko Ikeda and NTT
> >> >> people did:
> >> >> 
> >> >> http://lists.linux-ha.org/pipermail/linux-ha/2007-October/028388.html
> >> >
> >> > Good stuff, thanks you for pointing it out. And mr Ikeda and NTT for
> >> > sharing!
> >> >
> >> >> 
> >> >> I don't know why their code was not included in Heartbeat. This
> >> >> is an important issue, so it should get more attention.
> >> >
> >> > Agreed!
> >> 
> >> What can I do to make it included in Heartbeat? (we call it as "SF-EX")
> >> 
> >> I and my colleagues would be really happy if it is included
> >> as a standard component in Heartbeat and available for everyone.
> >
> > I'll try to test it and take a look at the code. There was also
> > some longish discussion on the list about it, which I didn't
> > follow very closely. Were there any technical issues unresolved?
> 
> I remember that the discussion point was that such functionality
> should be implemented as a quorum plugin from the Heartbeat design.
> 
> But as far as we studied, SF-EX can not be implemented as a
> quorum plugin because the quorum plugin is not resource-aware
> and therefore it can not guarantee to avoid a duplicate invocation
> of resources. Maybe we have just short knowledge of quorum though...

Not sure if I follow. You're worried about the plugin being
executed in parallel?

> So we believe that SF-EX can be included as it is
> as one of the options for protecting the data of resources.

Yes. Having it as quorum plugin would be great though.

> > Does anybody use sf-ex?
> >
> > There's another big contribution of yours in the queue: procd.
> > Since it depends on both the CRM and heartbeat and since the two
> > are not anymore in the same development tree we'll have to fix
> > the Makefile and most probably the include statements.
> 
> Thank you for remembering it. 
> Actually we're just working on updating procd now to adapt to
> Pacemaker and I'll post it very soon.

Thanks a lot.

Cheers,

Dejan

> 
> 
> Regards,
> 
> Keisuke MORI
> NTT DATA Intellilink Corporation
> 
> _______________________________________________
> 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