[Linux-HA] confusion of 'autojoin any' and 'ucast' directives

[Andrew Beekhof]

On 9/29/06, Brian Reichert <reichert at numachi.com> wrote:
> On Fri, Sep 29, 2006 at 11:34:46AM +0200, Andrew Beekhof wrote:
> > On 9/28/06, Brian Reichert <reichert at numachi.com> wrote:
> > >I'm exploring 'auto join' any as a way of simplifying cluster provisioning
> > >with CRM, but I'm confused by the apparent need to manually populate a list
> > >of peers for 'ucast' directives.
> > >
> > >Is there some means by which a node that is autojoining can communicate
> > >this to the CRM?
> >
> > only by using bcast or mcast.
>
> Generally, what I'm trying to address:
>
> I'm trying to build up a CRM-managed cluster that spans multiple
> netblocks.
>
> I've gotten the impression that the hostcache file is considered
> authoritive over ha.cf for what contains the list of nodes.   (Am
> I wrong with this?)
>
> I hoped that between autojoin, and adding/removing nodes via cibadmin,
> I could side-step the issue of ha.cf disagreeing with the hostcache
> file.
>
> I had thought that ucast would be my best decision for letting CRM
> traffic get to all of the nodes given our topology, but that
> re-imposes the constraint that I have to manage per-node details
> in ha.cf.
>
> Perhaps my assumption concerning ucast is inaccurate?
>
> I had considered using mcast, but don't know what the conventions
> are for describing an mcast group.  Still doing some research
> there...
>
> Anyway, of you (or anyone else) can offer some pointers (or
> corrections), that'd be great...
>

I believe the problem is that without the extra ucast directive(s) the
new node isn't yet able to speak to the other nodes for the autojoin
to even happen.

Maybe Alan is able to correct me if I'm wrong.