|
I have a cluster config for a master-master database, each with a slave:
Slave -- Master (primary) -- Master (secondary) -- Slave
I have a VIP for the primary master (read/write database) and another VIP
for load balancing (both slaves and the secondary master).
I'm running the directors on the master servers. Therefore, all servers
have lo:0 up.
Under version 1.2.3 and kernel 2.6.9-42.0.3 (redhat) things worked fine.
Switching between the masters lo:0 seem to be preserved (I think).
Therefore, load balancing worked on all 3 slaves (2 slaves, secondary master).
However, in version 2.0.7 and kernel 2.6.9-42.0.8, lo:0 disappears OR the
VIP for the read server (eth0:1) does not get created if lo:0 exists. If
lo:0 doesn't exist, eth0:1 gets created like it should. Both lo:0 and
eth0:1 have the same IP (load balancing IP), therefore, it looks as though
the VIP conflicts with lo:0.
I don't know if the configuration was correct under 1.2.3, but it worked
flawlessly for over a year. Under version 2.0.7, there is no failover since
the VIP doesn't get created for the load-balance IP or I have to manually
start lo:0 (that's not failover!).
Haresources:
<host> IPaddr::10.0.3.17/24/eth0
<host> IPaddr::10.0.3.20/24/eth0
LDirector:
virtual = 10.0.3.20:3306
real=10.0.3.13:3306 gate 12 ?
real=10.0.3.21:3306 gate 13 ?
real=10.0.3.23:3306 gate 12 ?
fallback=127.0.0.1:3306 gate 1
service=http
checktype=negotiate
protocol=tcp
scheduler = wrr
lo:0:
DEVICE=lo:0
IPADDR=10.0.3.20
NETMASK=255.255.255.255
NETWORK=10.0.3.0
BROADCAST=10.0.3.255
ONBOOT=yes
NAME=loopback
What's different? Kernel? Heartbeat? Is there another config option to make
it work?
Thanks.
|
