|
On Sun, 2007-07-22 at 22:24 -0400, Gerry Reno wrote:
> What exactly would be the benefit of notify_master/backup/fault and this
> script. It just puts a VIP/32 on 'lo' for the backup director and delete
> the VIP/32 from 'lo' for the master director. Is this necessary? I mean
> wouldn't you just need to let keepalived move the VIP/24 from one
> director to the other and not worry about anything about VIP on 'lo' for
> directors?
Remember: LVS is simply an intelligent router. Unless you choose to use
NAT, the realservers *must have the VIP* on them in order for a
listening process to handle connections.
In your case, all LVS is doing is saying "next hop for $VIP is
realserver N". Realserver N then needs to actually do stuff with the
incoming packet.
Note that it is possible (in theory if not in practice) to use iptables
and fwmarks to achieve the same result, which in turn means that you
don't need to have the VIP on the realserver. This is, however, more
complex - but in your specific case, where your realservers are also the
directors, can solve a number of associated problems which only occur
when using localnode. Please look back through the ML archives for
recent posts from me describing how to achieve this by using MAC
filtering within an iptables ruleset.
> To test I am using the 'rr' scheduler and sure enough the accesses are
> bouncing from one real server to the other. But this is occurring even
> when I am in a session. Is there no session 'stickiness' with the
> schedulers?
Not within the schedulers, no - you need to use persistence. "man
ipvsadm" and "man keepalived.conf" will be your friends here :)
Graeme
|
