all clients disconnect on failover

[Sebastian Vieira]

I've removed heartbeat/ldirectord and installed keepalived, in hope that the
failover works now. Unfortunately it doesn't. A manual failover (stopping
keepalived) works as expected, with clients staying connected to the
realservers, but when an unexpected failover occurs, they all disconnect.
I've followed the howto, read the docs/manpages and google'd for this, but i
can't find anything useful.

The method i've used is LVS-NAT. What am i doing wrong here?

My keepalived.conf:

! Configuration File for keepalived
global_defs {
   notification_email {
	sebastian at local
   }
   notification_email_from LVS05
   smtp_server 192.168.50.80
   smtp_connect_timeout 30
   router_id LVS05
}

! vrrp sync groups
vrrp_sync_group ZISLVS {
	group {
		VI_1
	}
	smtp_alert
}

! vrrp instances
vrrp_instance VI_1 {
	state BACKUP
	interface eth0
	track_interface {
		eth0
		eth1
	}
	nopreempt
	debug
	lvs_sync_daemon_interface eth1
	virtual_router_id 66
	priority 150
	advert_int 1
	authentication {
		auth_type PASS
		auth_pass ****
	}
	virtual_ipaddress {
		192.168.50.110
	}
}

! virtual servers
virtual_server 192.168.50.110 23 {
	lb_algo wlc
	lb_kind NAT
	protocol TCP
	ha_suspend
	persistence_granularity 255.255.248.0
	! real servers
	real_server 192.168.14.13 23 {
		weight 1
		inhibit_on_failure
		TCP_CHECK {
			connect_port 23
			connect_timeout 60
		}
	}
}


Kind regards,

Sebastian


On 10/5/06, Sebastian Vieira <sebvieira at gmail.com> wrote:
>
> Hi,
>
>
> I'm using LVS 1.2.1 in combination with Heartbeat v2.0.7 and ldirectord
> v1.77.2.39. When i do a manual failover (stopping heartbeat on the primary
> node) almost all connections get transferred perfectly to the backup node.
> But when i have an unexpected failover (eg: i yank the powercord from the
> active node) all clients are disconnected.  I've played with all the
> settings concerting keepalive/warntime/deadtime and have brought it down to
> the absolute minimum:   keepalive 30ms / warntime 75ms / deadtime 100ms.
> Anything lower than dead will result in alot of errors from heartbeat. The
> connections are simple telnet ones. I've ran out of ideas :(
>
> Anyone with a bright idea?
>
> Thanks, and kind regards,
>
>
> Sebastian
>