[Linux-cluster] Piranha question: Running client on a real serverhangs

[Lon Hohberger]

On Wed, Apr 25, 2007 at 10:56:24AM -0500, Brooks, Jody wrote:
> Thanks for the links... working my way through them now ...
> 
> For info: I am using the redirect bits like in "method #2".
> 
> In fact, that's all I'm doing on the real servers as I have not assigned
> the VIP to the lo interface on the real servers as seems to be discussed
> in LVS-DR.html (haven't read all that yet but that appears to be implied
> as I've read that in other places as well).  This VIP on lo doesn't
> appear to be critical as my service works fine when the client is on a
> non-real-server/non-director host.  Am I missing something here?  Seems
> like the VIP on lo would be used if you weren't doing the iptables
> redirect (i.e., method 2)... right? wrong?  

That's right; the redirect is a completely different solution to the
"ARP Problem" that LVS-DR.html talks about.  You need one or the other,
but not both.

What I was saying is that if you did #1 in the doc (or configured the IP
on lo as per LVS-DR.html), the connection would always work - it just
wouldn't be load balanced.  That would be a way to get rid of part of
your failure cases, if elimination of the failure was more important than
being optimally load balanced.

> I will try to incorporate some of the pieces from the LVS-DR.html ideas
> and see if that gets me any further.  

> For the record, our need of this is that one of our services is also a
> consumer of other of our services, all of which should be load-balanced
> across the same cluster as they're related.

If you can't get it working with LVS, you could always try something
like RRDNS to do something similar.

Unfortunately, I have not tried this configuration at all - so my
insight will naturally be rather limited. :(

-- 
Lon Hohberger - Software Engineer - Red Hat, Inc.