all clients disconnect on failover

Sebastian Vieira sebvieira at
Tue Oct 17 14:13:14 BST 2006

Well, i'm happy to report that everything is working after the gateway was
changed to reflect the vip :) I also switched back to the
heartbeat/ldirectord combination instead of keepalived, because i find the
administration a little better. But i guess that's a personal preference. I
also like to 'see' the interface (eth0:0 in my case) on the active node when
i do an ifconfig. Something which keepalived handles in some other way that
i don't quite get. I also find the development of ldirectord a bit better
than keepalived. Again, personal preference i guess.

In the next couple of days i'm going to put some clients on the cluster ip
and do some (unannounced) failovers to see what comes up. If all goes well
(and i don't see why it wouldn't) i'm going to contact the maintainer of the
realservers again and have them switch the rest of them to the gateway of
the director.

Thanks for all the help!


On 10/12/06, Sebastian Vieira <sebvieira at> wrote:
> Hi,
> Thanks everyone for the replies. I realize by your postings that there
> simply is no other way than changing the gateway on the realservers. Joe's
> option of plugging the realservers into the back of the director is (i
> think) not an option for us, as they are physically spread throughout the
> building. But maybe i understand you incorrectly.
> Brett: i definately want the benefits from LVS :)
> I'll contact the people who're maintaining the realservers and drop the
> request there. As a funny sidenote: I quickly checked one of the servers
> yesterday and it's running linux-2.0.28 :)  very stripped-down because
> even simple commands like ping are removed from it. Makes my work a whole
> lot ... uhm, more interesting.
> I'll get back as soon as i have an answer. It would be great if i could
> get this working.
> Thanks for all the help so far.
> Kind regards,
> Sebastian

More information about the lvs-users mailing list