[Debian-ha-maintainers] Bug#600095: Bug#600095: cman tells corosync to bind to 127.0.0.1 by default in squeeze

[Guido Günther]

On Thu, Oct 14, 2010 at 10:19:51AM +0100, Tim Small wrote:
> On 13/10/10 21:33, Guido Günther wrote:
> >On Wed, Oct 13, 2010 at 05:45:46PM +0100, Tim Small wrote:
> >>Package: cman
> >>Version: 3.0.12-2
> >>Severity: important
> >>
> >>By default cman tells corosync to bind to 127.0.0.1, and doesn't
> >>document how to get it to do otherwise (I'd guess because it's doing a
> >>lookup on the nodename, and somehow ending up with 127.0.0.1 - maybe
> >>this is different from the default RHEL config).
> >>
> >>Some sort of working-out-of-the-box example cluster.conf would seem
> >>essential if anyone wanted to get this package to work without screaming
> >>a lot.
> >>
> >><cluster name="fish" config_version="1">
> >>         <clusternodes>
> >>                 <clusternode name="squeeze-test" nodeid="1">
> >>                         <multicast addr="224.0.0.1" interface="eth0"/>
> >>                 </clusternode>
> >>
> >>                 <clusternode name="squeeze-test2" nodeid="2">
> >>                         <multicast addr="224.0.0.1" interface="eth0"/>
> >>                 </clusternode>
> >>         </clusternodes>
> >>
> >>         <cman two_node="1" expected_votes="1">
> >>         </cman>
> >></cluster>
> >This looks more like a bug in your configuration. Please make sure your
> >hostname doesn't resolve to 127.0..1.1 or similar.
> 
> That is the default in Debian, isn't it, and doesn't seem to cause
> any problems for other applications...
> 
> If there's no way (or no documented way) to bind cman to a
> particular interface, the it will also surely fail in the common
> case of having a public interface and a private interface?
> 
> In the case where:
> 
> <multicast addr="224.0.0.1" interface="eth0" />
> 
> is specified, then surely cman should use the primary IP address for
> the eth0 interface, rather than 127.0.0.1 - binding to anything
> under 127.0.0.0/8 is always going to be the wrong thing to do
> surely?
> 
> At the very least, if /etc/hosts needs to be modified from the
> Debian default in order for cman to work correctly then this should
> be documented, no?
Yes, documenting this would certainly be good. Care to send a patch
against README.Debian?
Cheers,
 -- Guido