Re: [LARTC] 1+1 HA gateway

[Mike Fedyk <mfedyk@matchmail.com>]

On Fri, Jul 13, 2001 at 09:53:24AM +0200, RoMaN SoFt / LLFB !! wrote:
> On Thu, 12 Jul 2001 11:20:18 -0700, you wrote:
> 
> >>  The situation I'm looking for is having two real routes to the
> >> destination network (via the fast gateway and the slow one
> >> respectively) but only the first (úst) one is used in normal
> >> conditions. The second (=slow) one will only be used in case the first
> >> breaks (i.e. failover mode).
> >
> >I believe the only way for the kernel to recognize that there has been
> >a failure, is the ethernet card detecting a line drop.  If you can be
> >sure that when the link goes down that this happens, you won't need
> >anything else except for the right rules in your routing setup.
> 
>  I forgot to say that I also tried to use metric: two routes with
> different metric (one with default metric [0, isn't it] and the other
> one with metric 10). This time I could enter the two  routes to the
> same destination but when the first gateway lose connectivity it seems
> not to be detected. I'll keep on doing some testing.
> 
>  I read in this list's archives that the kernel routing code should
> detect if the gateway is ok and in negative case switch to another
> route (with greater metric). How does it work exactly? Which type of
> checks are performed?
>

I don't really know.  I haven't seen any traffic on the network that
seems to try to detect connectivity.

>  Mike, the issue is not to detect when ethernet is broken (this is a
> feature of the network card and it is used, for instance, in "bonding"
> driver; indeed in that case my problem would be solved using this
> driver in backup mode) but detecting when the destination network is
> not reachable. So the gateway itself could be ok (it could have its
> "receiving" ethernet up, I mean, my linux router [which I'm trying to
> config] can reach the gateway) but its output line could be down
> avoiding a correct deliver of packets (gateway can reach destination
> network).
> 
>  Is it absolutely necessary to use a routing daemon in my case? Or the
> metric trick should be sufficient for me?
> 

I would guess that the kernel would need some icmp message sent to it
to detect that a route is down.  Like "dest unreachable" from the
first hop.

In my case, I have a bridged DSL connection, and if the link goes
down, I won't get the icmp either.  I have another routed dsl
connection, but I haven't tested with that yet.  With the bridged
line, the packets go out, and don't get any response.  The kernel
doesn't do anything in this case.

Do you know anything about C or C++ coding?  If so, you could take a
look at the routing code yourself and maybe get an idea of what is
going on.

My guess is that it requires dest-unreach to work.  It'd change in the
routing cache, and you wouldn't see anything in your other tables
change.

Mike


_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/