|
Hello,
On Fri, 9 Jun 2000, Jeremy Hansen wrote:
>
>
> Well there is error handling with dns, infact it's pretty much fundamental
> feature of dns to have secondary name servers so how would that hurt
> things....or are you speaking of a lvs machine going down and rr pointing
> to a downed lvs server? I see what you mean there but I would still
> institute lvs failover as I would without using rr dns, so behind the dns
> it would really be the same type of setup. It just seems that rr dns
> would be a very simple way to balance across multiple lvs machines.
>
> You could have real servers point to all lvs machines somehow, I think
> fwmark can be used here, or have a unique group of real servers per lvs
> machine. It would be neat I think, but is there real world benefits to
> doing this?
My thoughts was about such setup (not tested):
nslookup www.<domain>.com
Name: www.<domain>.com
Addresses: 192.168.1.100 192.168.2.100
Your auth name servers:
192.168.1.2
192.168.2.2
ISP 1 ISP 2
\ /
192.168.1.100 \ / 192.168.2.100
LVS
The result:
- Packets for 192.168.1.100 and 192.168.2.100 are marked
with same value using ipchains -m
- the virtual service is defined via ipvsadm -f
- if one ISP is down you can update the state in the other
DNS, you can use little timeouts too. Is TTL=5 minutes a
problem? May be for some proxy servers?
- you handle one virtual service
- we assume you can replace the director if there are
problems, you can use as many backup servers as you wish
- we are happy if both ISPs work :)
OK, this is only an example setup. You still can use
many LVS boxes which can use same real servers without
fwmark.
>
> I kind of agree with you on the hackish thing of using rr dns, but dns is
> weird I think. There are a lot of thing about dns that don't necessarily
> follow standard yet people still do it and bind is written to allow such
> behavior. Hey, if it's good enough for kernel.org, it's good enough for
> me.
>
> -jeremy
Regards
--
Julian Anastasov <uli@xxxxxxxxxxxxxxxxxxxxxx>
|
