|
Hi,
I'm trying to implement a two box lvs cluster with failover.
I'm very close to having this working. I have heartbeat installed
and configured to use both lans with UDP. I test that the VIP (10.0.3.3)
switches when I disconnect either server. They switch on a VIP of eth:0.
I then moved on to the load balancing portion.
This is what I have:
client 10.0.3.5
|
|
------------------------- 10.0.3.0/24
| |
eth0 | 10.0.3.1 | 10.0.3.2
box 1 box 2
eth1 | 192.168.1.1 | 192.168.1.2
| |
-------------------------- 192.168.1.0/24
(cross-over cable)
Box 1 and 2 are both directors and servers.
I'm using the latest 2.4.X patches (Apr 20th) and ipvsadm 1.17
I installed the following on both boxes:
ipvsadm -A -t 10.0.3.3:80 -s rr
ipvsadm -a -t 10.0.3.3:80 -r 10.0.3.1:80 -g -w 1
ipvsadm -a -t 10.0.3.3:80 -r 10.0.3.2:80 -g -w 1
I have apache listening on 10.0.3.3 and setup different
index pages so I could tell which real server I was hitting.
I fired it up and it kinda works. It will switch back and forth
a few times then hang. If I hit stop it will then work again
a few more times. Looking at the activity LEDS on the hub I noticed
they were on solid between the two boxes.
I read as much as humanly possible of the HOWTO and figured it might
have something to do with the "arp problem" so I added the "hidden"
patch to both kernels, changed the VIP for the real servers to 10.0.3.4
(left it 10.0.3.3 for the Directors/HeartBeat), and ifconfig'd dummy0
on both boxes to 10.0.3.4, and turned hidden on in /proc/sys/net/...dummy0
I still get the packet storm.
I'm going to try to sniff the network and see what's going on but I'm
rapidly approaching areas where I'm a total novice.
I think there must be something stupid I'm doing ?
Thanks,
Doug
|
