[j-nsp] Strange ARP issue on M7i

[Jonathan Lassoff]

On Tue, Aug 14, 2012 at 1:00 PM, Tobias Heister <lists at tobias-heister.de> wrote:
> Hi
>
> Am 14.08.2012 15:12, schrieb Markus:
>> Isn't that weird? Where did that arp entry come from and why was it saved on the Juniper for so long, and only got removed after I removed the static routing of that /24?
>
> We saw a similar thing a short time ago on an MX480 running 10.4R9
> In our case it was a bgp route pointing to a no longer existing ip address as the next-hop. The arp entry for this ip address stayed active as long as there was an active route for it.
> Even clearing the arp cache witch clear arp hostname x.x.x.x did not do the trick. The next-hop ip was gone for several weeks and the arp entry had low timeout values left but never expired.
> After clearing the route the arp entry vanished as expected.
>
> I guess something keeps the arp entry from being deleted as long as there are or were forwarding entries in the fib for it at any time.

Probably because the underlying information ARP is learning is used to
build the next-hop in the forwarding table (which needs to know what
Ethernet address to put in the destination MAC).

However, I would think that the route should become unreachable or
pruned if ARP is failing.
What if the remote router died for some reason? If the ARP entry and
next-hop were kept into place, the path would not work, but the route
would stay active.
A dynamic routing protocol and BFD would be see this right away and
move traffic, but this would break any static routes that rely on any
dynamism with ARP and next-hops.

Moral of the story, as I see it: avoid static routing.

--j