[Pacemaker] Stonith Device APC AP7900

[Dejan Muhamedagic]

Hi,

On Wed, Nov 17, 2010 at 05:01:20PM -0600, Andrew Daugherity wrote:
> > Message: 3
> > Date: Tue, 16 Nov 2010 11:24:26 +0100
> > From: Dejan Muhamedagic <dejanmm at fastmail.fm>
> > To: The Pacemaker cluster resource manager
> > 	<pacemaker at oss.clusterlabs.org>
> > Subject: Re: [Pacemaker] Stonith Device APC AP7900
> > Message-ID: <20101116102426.GA22675 at rondo.homenet>
> > Content-Type: text/plain; charset=us-ascii
> > 
> > Hi,
> > 
> > On Mon, Nov 15, 2010 at 10:41:22AM -0700, Devin Reade wrote:
> > > --On Monday, November 15, 2010 08:40:45 AM -0700 Rick Cone
> > > <rcone at securepaymentsystems.com> wrote:
> > > 
> > > > In production I am planning to have 2 separate AP7900 units each plugged
> > > > into 2 different APC UPS units to achieve that.  I would then have the
> > > > single node name on each, for each of the 2 PS's on the individual
> > > > systems.
> > ...
> > 
> > Right, there's currently no way to do a simultaneous reset on two
> > distinct fencing devices.
> > 
> > > I think that in such a case you'd also have to ensure that your stonith
> > > action is poweroff rather than reset, or your node may not actually
> > > lose power (although you could mitigate that likelihood by configuring
> > > a longer reset time in the PDU).
> > 
> > Defining more than one stonith resource wouldn't work in this
> > case either, because as soon as one of them reports success, the
> > node is considered fenced.
> 
> There is one possible solution to this -- the APC switched PDUs do
> support being configured in multicast groups, and you can configure
> groups of outlets together, even spanning several PDUs.  For example you
> could say PDU 1, outlet 1 and PDU 2, outlet 3 are known as "webserver1",
> and then tell it to "power cycle webserver1".  If it's not already
> supported by the stonith agent it shouldn't be too hard to add it --
> APC's syntax is actually documented fairly well.

There's a mention of that in README.rackpdu. I was even
castigated for adding it :) For, apparently, the feature can
fail without giving the caller any error indication. At least
that's what somebody reported.

Thanks,

Dejan

> The obvious disclaimer is that I haven't tried this myself.  I do have a
> similar setup, but decided it was easier to use the IPMI stonith agent
> (external/ipmi) to control my servers.
> 
> -- 
> Andrew Daugherity <adaugherity at tamu.edu>
> Texas A&M University
> 
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker