LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: IPVS Synchronization bug

To: Fabrice <fabrice@xxxxxxxxxx>
Subject: Re: IPVS Synchronization bug
Cc: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
From: Julian Anastasov <ja@xxxxxx>
Date: Fri, 26 Oct 2001 11:37:22 +0300 (EEST)
        Hello,

On Fri, 26 Oct 2001, Fabrice wrote:

> Hello,
>
> I made some tests today:
>
>
> With "ip route":
> ================
>
> I added the following rule to the master load balancer:
>
> ip route add 224.0.0.81 dev eth2 src 192.168.3.1

        Right. If you don't have specific route for the used
multicast address(es) then the kernel uses your default route.

> And it worked (you need to restart the synchronization
> daemon on the master load balancer).
>
>
> With "route":
> =============
>
> I executed the following command:
>
> route add -host 224.0.0.81 dev eth2

        Yes, source address is selected from eth2. I assume in
your previous tests you didn't used such route?

> And, as Julian told me, I saw that it didn't set the "prefered
> source address" parameter in the routing table. But even
> without it, it did work because it took the IP address of
> eth2 (which is correct).
>
> Btw, the routes seems to have been configured with "ip route"
> because they all have their "prefered source address"
> parameter correctly filled. Maybe the "route" command sets
> it by default to the interface's address.

        Yes, for directly connected networks the routes added
by kernel (proto kernel) when IP address is added contain this
address as preferred source. But when other routes are added (proto
boot) the user should specify the prefsrc.

        So, you report that even 0.9.4 with correct route selects
the right source address? I'm asking that because I have never tried
it (syncd) :)

> Thanks,
>
> Fabrice Bucher


Regards

--
Julian Anastasov <ja@xxxxxx>



<Prev in Thread] Current Thread [Next in Thread>