Hi Claudio,
just finished building it all and it seems it works great, very impressive.
Thank you!
I'll be running a few tests tomorrow to see if it's stable on the lab, I'll
keep you posted.

BTW when testing some network statements I think I might have hit a bug.

A.A.A.A is a connected route
B.B.B.B is a static route (route -T 99 add B.B.B.B/28 3.3.3.3)

Config:
rdomain 99 {
                rd 1.1.1.1:99
                import-target rt 1.1.1.1:99
                import-target rt 2.2.2.2:99
                export-target rt 1.1.1.1:99
                depend on mpe99
                network A.A.A.A/28
                network B.B.B.B/28
}


With this config, on PE2 I never get any BGP updates for B.B.B.B (even though
it's marked as announced on PE1).

Now I edit the config :
network B.B.B.B/28 set nexthop self
+bgpctl reload

Route is now announced and I get the update on PE2, everything ok.

I reboot the machine with the exact same config, route is gone on PE2. I edit
the config (INTFADDR being the same address than with self)
network B.B.B.B/28 set nexthop INTFADDR
+bgpctl reload

Route gets to PE2.

Reboot ... Same problem. I switch back to self and the routes comes in again.

The problem is easily reproducible and occurs every time (reboot or process
kill).

Any ideas?

Thanks,

Rimi

Le 7 avr. 2012 ` 22:02, Claudio Jeker a icrit :

> On Wed, Apr 04, 2012 at 10:08:48PM +0200, Rimi Philippe wrote:
>> Hi Claudio,
>> It works at 90% thanks.
>>
>> The last 10% are still not working. On PE1 I have 2 Rdomains (20,30) and
PE2 1
>> rdomain (20). On PE1 I want the rdomain 20 routes to be imported in rdomain
30
>> (locally), but that doesn't seem to work locally, here are the details.
>>
>
> The last 10% are the hardest ones. The following kernel and bgpd diffs
> should do the trick. The kernel diff is needed to forward traffic
> directly between mpe(4) interfaces. The bgpd diff makes sure that the MPLS
> routes are inserted into the other VPNs on the same router.
>
> This works for me but is not heavily tested please test.
> --
> :wq Claudio

Reply via email to