[ULTRAMONKEY-USERS] UM3 & High Capacity Highly Availability and Load Balancing problem

[Nathan Brown]

That was my initial reaction as well, I first put the following in my fedora
/etc/sysctl.conf (remember that the fedora boxes run with eth1):

net.ipv4.conf.all.arp_ignore = 1
net.ipv4.conf.eth1.arp_ignore = 1
net.ipv4.conf.all.arp_announce = 2
net.ipv4.conf.eth1.arp_announce = 2
 
when that didn't work I used arptables_jf with the following:
/sbin/arptables -A IN -j DROP -d 192.168.2.10
/sbin/arptables -A OUT -j mangle -o eth1 -s 192.168.2.10 --mangle-ip-s
192.168.2.14
 
Would this really be an ARP problem or maybe the fact that im using debian
etch with the original UM3 install?
 
Would updating heartbeat and ldirectord help at all in this situation?

 

  _____  

From: ultramonkey-users-bounces at vergenet.net
[mailto:ultramonkey-users-bounces at vergenet.net] On Behalf Of Glen Kendell
Sent: Tuesday, June 26, 2007 9:41 PM
To: ultramonkey-users at ultramonkey.org
Subject: Re: [ULTRAMONKEY-USERS] UM3 & High Capacity Highly Availability and
Load Balancing problem

 

maybe you've already answered your own question.  assuming you are using
direct routing, what have you done on the fedora boxes to suppress arp?

On 6/26/07 3:41 PM, "Nathan Brown" <nathan at fishflirt.com> wrote:

I have 2 load balancers setup running debian and 2 web servers setup running
fedora 7
 
My main problem is that when the load balancers first come up and start
routing traffic everything works fine, I can view ipvsadm and see that it
can see both web servers just fine.  The problem comes about 20 or 30
seconds after the the load balancers starts routing.  What happens is one of
the web servers drops and then a few moments later the other web server
drops and then it adds the fail back 127.0.0.1 to the list being the only
active server.
 
What I cant figure out is why for those first 20 or 30 seconds I can use
everything wonderfully but after that some catastrophe happens.  Is it an
ARP issue with fedora?
 
I was able to use the debian setup for the real servers without a problem,
sysctl.conf took the arp setting without error.  The only other thing that
might throw it I guess would be that my balancers operate on eth0 and the
web servers operate on eth1
 
Let me know if anyone has any suggestions

  _____  

_______________________________________________
Ultramonkey-users mailing list
Ultramonkey-users at vergenet.net
http://lists.vergenet.net/listinfo/ultramonkey-users





-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.vergenet.net/pipermail/ultramonkey-users/attachments/20070627/c94fa8ac/attachment.htm