LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: ARP reply control

To: David Osborne <David.Osborne@xxxxxxxxxxxxxxxx>
Subject: Re: ARP reply control
Cc: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: Julian Anastasov <ja@xxxxxx>
Date: Thu, 19 Jul 2001 20:11:41 +0300 (EEST)
        Hello,

On Thu, 19 Jul 2001, David Osborne wrote:

> On Mon, Jul 16 2001 at 19:36:13 +0300, Julian Anastasov wrote
> >     Did you tried http://www.linux-vs.org/~julian/route-noarp.txt ?
> > There is a section "Notes for the LVS users".
>
> Yes, but it didn't help me.
>
> On Mon, Jul 16 2001 at 20:04:17 +0300, Julian Anastasov wrote
> >     About the used device: check whether you can switch to using
> > rp_filter for your devices (all):
> >
> > echo 1 > /proc/sys/net/ipv4/conf/all/rp_filter
> > echo 1 > /proc/sys/net/ipv4/conf/eth0/rp_filter
> > echo 1 > /proc/sys/net/ipv4/conf/eth1/rp_filter
> >
> >     If not, try the same with "arp_filter".
>
> I tried both, but didn't help.

        Hm, it depends on the used routes. They really solve problems
when the routes a properly configured.

> >     In Linux 2.2+ all devices respond to remote ARP probes no
> > matter the requested address is local or remote. In Linux, by placing
> > one IP on specific device you can achieve only:
> [...]
> >     I don't remember for other reasons one to place an IP address
> > to specific device. So, placing VIP on "unused" eth card (eth1) does
> > not solve the problem for the announced MAC. All devices reply to all
> > broadcast ARP probes until some policy is applied:
>
> The system acting as LVS director has 3 interfaces; I was attempting
> to use one for heartbeat with the backup using a crossover cable and
> to use the other two to share traffic -- this was probably my mistake.
>
> I've since given up using all three interfaces and am currently trying
> one connected to the network and one as private (10.0.0.x) with the
> crossover to the second system, with the third unconfigured. That
> seems to work: the Windows clients no longer seem to be confused by
> ARP replies variously giving one or other of the MAC addresses of the
> two publically-connected interfaces.

        Then try to use the hidden flag instead of the per-route arp
flag. It has longer success history.

> My only problem left is that although I can ping one director from the
> other over the crossover cable, and make a telnet connection, I can't
> use it for heartbeat. If I start heartbeat on the backup director
> using that interface for the udp heartbeat, it doesn't see a reply and
> starts attempting address takeover.

        Hm, may be if you export some settings the heartbeat gurus can
help you :) Is it related to some broadcast requirements? To solve the
ARP problems the routing structure will be useful. I see that you use
non-traditional setup but I can't imagine everything. I hope you can
use pseudo network numbers to express your topology. There are so many
guys on this list that after a first look can tell you where is really
the problem. You'll be surprised :))) It saves time when the settings
are shown in the early postings.

> --
> David Osborne                         david.osborne@xxxxxxxxxxxxxxxx
> Academic Computing Services     phone/voicemail: +44 (0)115 951 3397
> The University of Nottingham                fax: +44 (0)115 951 3358
> Nottingham NG7 2RD, UK          http://www.nottingham.ac.uk/~cczdao/


Regards

--
Julian Anastasov <ja@xxxxxx>



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