all clients disconnect on failover

[Graeme Fowler]

On 10/10/2006 14:53, Sebastian Vieira wrote:
> Okay, i've done that (windump instead of tcpdum, but ok), and when i
> powerdown the active node i see the following message repeated alot:
> 
> 15:29:22.231586 arp who-has 192.168.50.110 tell 192.168.50.110
<snip>
> Funny thing is, 192.168.50.110 is the cluster ip. I'm not too familiar with
> tcp (yes, i'm learning) so i don't know what this means. No more packets to
> the client either are sent, so the connection just times out (did i say 
> that correctly? :))

If you're using keepalived (I'm sure you said you were), then these are 
the gratuitous ARP (GARP) packets being sent out to refresh everyone's 
ARP cache with the MAC of the new director.

I have a theory here, though... you're doing NAT. Therefore the director 
is the default gateway for the realservers - is the address of their 
default gateway the same as the VIP?

If not, you need another vrrp_instance defined which makes the default 
gateway fail over to the backup director.

Graeme