Mailing List Archive: Ipvs with NAT ...

Hi,
I've got a routing problem.
I'm working with lvs 0.1.2 and kernel 2.4.0 ...

My net work seem to be like that :

------
|_________FW____________Load-Balancer_____________Real Web Server
Net | |_________________________________________|
------

The Real server shown above has his default gw set to the FW box
When asking the virtual server IP for a connection, the real server receive
the packet, but return its result
to the source IP by the FW, not via the LoadBalancer box.

Do you know if there's a way to set automaticaly the return route to the
Load-balancer in case of a source
range IP ?
Explicitly, the principe is that one IP, IP1, must pass through the FW,
instead of IP2 that mustn't ...

PS: Sorry for my poor english ...

Regards,

--
Lifo.

Hello,

On Tue, 30 Jan 2001, Sebastien COUREAU wrote:

> Hi,
> I've got a routing problem.
> I'm working with lvs 0.1.2 and kernel 2.4.0 ...
>
> My net work seem to be like that :
>
>
> ------
> |_________FW____________Load-Balancer_____________Real Web Server
> Net | |_________________________________________|
> ------
>
> The Real server shown above has his default gw set to the FW box
> When asking the virtual server IP for a connection, the real server receive
> the packet, but return its result
> to the source IP by the FW, not via the LoadBalancer box.

The route in the real server to FW must point to the LB box.
But sometimes this is not enough. You must stop the ICMP redirects
in the LB box, echo 1 > conf/*/send_redirects

> Do you know if there's a way to set automaticaly the return route to the
> Load-balancer in case of a source
> range IP ?
> Explicitly, the principe is that one IP, IP1, must pass through the FW,
> instead of IP2 that mustn't ...

You have to read the threads from the last week in the mailing
list. There was a discussion about how to run NAT on same LAN.

LVS 0.1.2 does not work with NAT-ed real servers when you put
ipfw or iptables NAT rules. You need to upgrade to 0.2.3 for this.

> Regards,
>
> --
> Lifo.

Regards

--
Julian Anastasov <ja@ssi.bg>