Hi Nenad,
i got keepalived to failover a ip-adress from A to B,
and back again when A starts up.
So far so good -
but it does not loadbalance the requests.
when the IP is on A then i only get responses from A,
B shows some entries with ipvsadm - but it looks like it thinks it is the
director
and vice versa.
No :) This another point I need to document :
Imagine the follwoing scenario :
o You have built an active/passive configuration for your LVS cluster
o Host A is MASTER and host B BACKUP
when you run Keepalived on both and run 'ipvsadm -ln' then you will see
that LVS topology appear on both director. :) Ones can say : "Hey it is not
working since I see both configuration on both director even on the BACKUP...".
Then software design implemented here is to speed up IP takeover for A to
B. Since VIP is only owned by MASTER director (A in normal state) then all
LVS topology will be only accessible from A director. So if this topology
is set on B it is not a problem since all traffic will be directed on host
A. But, to minimize IP LVS availibility during VRRP IP takeover, then it is
better if lVS rulez are already set :) then the keepalived framework will
only activate healthcheckers for this VIP.
So if VIP is on A in normal state then LVS topology will appear on A & B
(normal, implementation choice). Since VIP is not on B then healtcheckers
on B will not be activated to not overload the realserver since it is not
needed. If A fire then B become MASTER and VIP is now owned on LVS B then
healtcheckers are activated. VIP presence drive healtcheckers activity.
Is it enough to configure keepalived, or do i have to do something
additional ?
No need to configure any other tools.
Btw. Keepalived does not work with my RTL-8139 based Cards,
it only reports "ethx down".
I switched to some 3Com cards which seem to work fine.
Yes, since I don t have realteck NIC I can not debug this issue.
Best regards,
Alexandre
|