[j-nsp] mx-class units now advertisement management interface networks in BGP

[Doug Hanks]

It's working as designed.

Junos leaves the BGP advertisements in the hands of the operator. What you've done is created an export policy that just happens to match fxp0; this isn't Junos' fault.

If you want to advertise direct interfaces, but exclude fxp0, you could do something like this that you could cut and paste across N routers without having to modify (thanks Harry for confirming):

term block-fxp {
    from interface fxp0.0;
    then reject;
}
term 2 {
    from protocol direct;
    then accept;
}

From: Jo Rhett <jrhett at netconsonance.com<mailto:jrhett at netconsonance.com>>
Date: Thu, 27 Sep 2012 13:06:30 -0700
To: Harry Reynolds <harry at juniper.net<mailto:harry at juniper.net>>, dhanks <dhanks at juniper.net<mailto:dhanks at juniper.net>>
Cc: "juniper-nsp at puck.nether.net<mailto:juniper-nsp at puck.nether.net>" <juniper-nsp at puck.nether.net<mailto:juniper-nsp at puck.nether.net>>
Subject: Re: [j-nsp] mx-class units now advertisement management interface networks in BGP

Reply to Harry and Doug both since you mostly asked the same question.

On Sep 27, 2012, at 12:13 PM, Harry Reynolds wrote:
It might help if you posted your BGP export policy. IIRC, there is a no-readvertise flag available for a static but not aware of any inherent blocking of the advertisement of an fxpo address via BGP, more so if your export permits it.

To me it is a bug to advertise a route which you won't route packets for. Obviously it's your fault if you advertise a route and have a packet filter blocking packets -- the routing engine isn't responsible for this. But fxp0 is supposedly on its own routing fabric. I can't send packets in ae0 destined for something on the fxp0 network.

If a route visible in one routing engine was advertised out by another routing engine (with no route-sharing between them) this would be a bug, yes? Why isn't fxp0 treated the same way?

Finally, we have the same export policy on every node in our network. Having to break that out, and hand-tune every export policy to explicitly deny the fxp0 interface's routes is a lot of work with zero gain. If for some reason Juniper feels that it's important to someone somewhere to announce a route you won't accept packets for, why isn't there any easy method to disable this nonsensical, nonfunctional, nobody in their right mind would or could use it (non)functionality?

Obviously, a feature request for "protocol bgp { interface fxp0 { ignore; }}" would do the trick, but I struggle to believe that you've never seen this problem before, and you don't have a better way to prevent this behavior.

--
Jo Rhett
Net Consonance : net philanthropy to improve open source and internet projects.