[PDB Tech] allow empty IP field or not?

[Chris Caputo]

On Tue, 27 Dec 2016, Job Snijders wrote:
> On Tue, Dec 27, 2016 at 01:50:13PM +0000, Chris Caputo wrote:
> > Sometimes folks have the IPv6 flag enabled, but they aren't yet doing
> > IPv6 at a particular IXP.  If they fill out the IPv6 address, we've
> > seen participants with automated session configuration based on
> > PeeringDB try to ND for a router that isn't yet configured to handle
> > the ND.  This results in broadcasts/multicasts on the fabric which our
> > janitor must then chase down to get quashed.  Members then tell the
> > janitor that, hey we saw it in PeeringDB, and then the janitor has to
> > explain that in the particular case they are not yet doing IPv6...
> > 
> > When this happens with IPv4 it is no big deal since the arpsponge handles 
> > it, but for IPv6 it is inconvenient.
> 
> Are you saying that when someone enters the IPv6 address in PDB, but
> doesn't configure the address on their router... you end up with people
> sending ND for that address? I would say that is the desired behaviour.
> If you don't configure the address you should't list it on PDB.

Networks don't necessarily know that and so when they get an assignment 
from the IXP of both v4 and v6, they fill in PeeringDB accordingly, even 
well in advance of beginning their use of IPv6.  Naive networks don't 
realize that some networks actually configure peering sessions based on 
PeeringDB.

As a fabric janitor, I'll email members ND'ing for an IP that won't 
respond, requesting either passive or disable of session, and will also 
ask the other member to remove their addr entry in PeeringDB until they 
are ready.

> Keep in mind that currently you have to EITHER configure an IPv4 or IPv6
> address, OR both an IPv4 and IPv6 address.
> 
> So, if people aren't ready to do IPv6: don't fill in the address. If you
> are deprecating IPv4: only fill in the IPv6 address. However, we could
> consider it an error if you don't fill in any address.

Sounds good.  Just pointing out an operational consequence to 
broadcasts/multicasts on the IXP fabric.

Thanks,
Chris/SIX