|
Hello,
are you capturing the traffic? ( ie: transparent mode ) or do you have to
configure the IP in the brownsers?
For the first option, you have to use the fwmark thing, that I didn't got
work to balance a service living in the same box.
For the second option, I have several working implementations ( maybe not for
3128 port, but 80... really, no difference ). Using ldirectord + heartbeat.
It was quite easy. In the RedHat kernels, Fedora, or whatever, I did have
some caveats on the ARP thing.
BR,
Francisco Gimeno
-----------------
> Good evening:
> I decided to write to this list in the hope that somebody could help me
> determine where are my (of course several) mistakes.
>
> For a couple of weeks I've been trying to have working a 2 node LVS-DR
> "cluster" under a HA schema (ie. "active/active" configuration).
> This cluster serves a Squid web cache "resource" (that works fine on
> each node).
>
> I've used heartbeat + ldirectord software (2.0.5 versions) under Fedora
> Core 4, and didn't touch the provided kernel (it's as installed), and
> besides that I tryed to imitate the configuration steps as stated in
> http://www.ultramonkey.org/3/topologies/sl-ha-lb-eg.html
>
> What I can see is that failover messages are interchanged between 2
> nodes (for example when I stop heartbeat in one node).
> After that with ipvsadm I can see that the node supposed to be down has
> weight=0 (it's ok).
> But when I try to use Squid through the VIP, I cannot access any site in
> the web (connections are dropped?).
> And once I resume heartbeat in the "failed" node, everything works fine
> again.
>
> Does anyone else has configured an lvs for use with Squid?
> Can you share it?
>
> Perhaps I'm not using the right software for this... can you suggest a
> better software "combo"? (I mean other than ldirectord)
>
> TIA
>
> Best regards
>
> Ignacio
|
