LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: arping from wrong interface.

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: arping from wrong interface.
From: Jan Klopper <janklopper@xxxxxxxxx>
Date: Tue, 27 Sep 2005 14:26:53 +0200
Horms,

Thanks,
That works great, the routers now see the correct if for the vip, and thus don't broadcast the packages around anymore.

Thanks again for a great piece of work, and support!

greets
Jan

Horms wrote:

On Tue, Sep 27, 2005 at 10:18:28AM +0200, Jan Klopper wrote:
Horms,

I tried messing about with the -arp flag on ifconfig eth1, but that pretty much took the entire interface down.
As expected the failover server saw this hapening, and took over the VIP.

The -arp flag probably isn't what you want here.

Changing it back to ifconfig eth1 arp, restored the link, but the secundary server kept the ip.

Setting net.ipv4.conf.eth1.arp_ignore = 1
in /etc/sysctl.conf, didn't do much different. it took the the internal ip down again.
And removing it put the ip back online.

How should i have done this exactly?
I read:
http://www.ultramonkey.org/3/topologies/sl-ha-lb-eg.html
And figured out that i would go with the debian way of things, altough im on gentoo.

eth0 should abviously arp for the VIP, but doesn't, eth1 does so, but shouldn't. eth1 should only arp for its internal ip. (so setting -arp on eth1 doesn;t do the trick)

Setting

# When an arp request is received on eth0, only respond if that address is
# configured on eth0. In particular, do not respond if the address is
# configured on lo
net.ipv4.conf.eth0.arp_ignore = 1



tells me i could add this to eth1 to make it ignore arps for the VIP, since its not bound to eth1.

any more help? (don't want to tinker about all too much since i can have the machine go dark on me, its colocated)

If the VIP is on eth0, and you don't want it advertised ovr ARP on
eth1, then set:

net.ipv4.conf.eth1.arp_ignore = 1
net.ipv4.conf.eth1.arp_announce = 2



<Prev in Thread] Current Thread [Next in Thread>