[Pacemaker] Low cost stonith device

Dejan Muhamedagic dejanmm at fastmail.fm
Tue Oct 6 06:50:02 EDT 2009


Hi,

On Tue, Oct 06, 2009 at 10:55:11AM +0200, Johan Verrept wrote:
> 
> Hi Lars,
> 
> > Yes.
> > hostname (-s, -f) and uname differ in that
> > uname -n just does one syscall (uname),
> > and hostname (typically) does a gethostname(),
> > then a gethostbyname(), the result of the latter
> > heavily influenced by whatever is in your nsswitch,
> > resolv and other .conf, as well as what may be listed
> > in /etc/hosts, and the order of those entries ;)
> 
> I noticed. So comparing uname and hostname -s isn't a good idea. I
> feared that.
> 
> > Instead of doing clever guesswork,

I'd also be happy to avoid that.

> > surely those:
> > 
> > > 	http://www.aviosys.com/ip_power_9258hp.html
> > > > > PS: there are others too in this category of devices:
> > > > > http://www.server-rack-online.com/cw-2h2-c20.html
> > 
> > can be configured to use "arbitrary" names?
> 
> As I stated before, the label is limited to 15 characters. Those fill up
> fast. That is why I am doing the whole guessing in the first place.
> Using the short hostname seemed a better idea than just cutting off the
> first 15 characters of the full hostname. It is much easier to
> reconstruct the full name.
> 
> > so if your uname -n does contain dots (FQDN),
> > you use that as label in the power switch,
> > and if not, there you are.
> > 
> > You just have to get it right then,
> > and the RA does not need to guess.
> 
> I could just check the "uname -n" for dots in the RA. Still a guess, but
> a better one.

Yes, better, but still a guess :)

> Offcourse, I could just move away from the whole port detection and just
> require a parameter that maps the ports to the hostnames. I am not a big
> fan of that though.

Well, if we already have the option to use port-nodename mapping,
that should be preferable to increasing the configuration size.
Still, perhaps you could add a hostlist parameter like in most
stonith plugins. Then just use that instead of trying to guess
the right node name.

Thanks,

Dejan

> 	J.
> 
> 
> 
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker




More information about the Pacemaker mailing list