Re: [PATCH docs/gui-tests/manager/network/proxmox{-ve-rs,-perl-rs} v3 00/13] Add IS-IS protocol to fabrics

[Gabriel Goller <g.goller@proxmox.com>]

On 01.09.2026 10:58, Stefan Hanreich wrote:
> 
> 
> On 9/1/26 10:55 AM, Hannes Laimer wrote:
> > On 2026-09-01 10:33, Gabriel Goller wrote:
> >> On 01.09.2026 09:09, Hannes Laimer wrote:
> >>> On 2026-08-31 14:59, Gabriel Goller wrote:
> >>>> On 31.08.2026 14:27, Gabriel Goller wrote:
> >>>>> On 31.08.2026 10:48, Hannes Laimer wrote:
> >>>>>> Aside from these very tiny nits/questions, looks very good to me:
> >>>>>>  - could we have the same `network_type`-nob here that we introduced for
> >>>>>>    OSPF?
> >>>>>
> >>>>> Yes, makes sense. ISIS supports network-type broadcast (unlike openfabric) and
> >>>>> even multi-access (e.g. multiple devices behind a switch).
> >>>>
> >>>> Oh, no wait I won't be adding this. ISIS only has broadcast and point-to-point
> >>>> available, and the default is broadcast. We use broadcast when an interface ip
> >>>> is set and point-to-point when no ip is set.
> >>>>
> >>>
> >>> yes, but we only need a specified ip for v4 broadcast, v6 does not. so
> >>> unnumbered v6 broadcast is not possible currently
> >>
> >> Hmm that's true. v6 broadcast works in two scenarios: link-local addresses
> >> (which IMO is not "unnumbered" -- but this is debatable :)), and because it uses
> >> ndp instead of arp, which doesn't have the weird address-selection.
> >>
> >> I thought a while back about this -- we don't need to duplicate the ipv6
> >> addresses to the interface. But then we should also remove the link-local
> >> addresses to make it clean.
> > 
> > i don't think we can, the ll would be the next-hop..
> 
> My thoughts exactly, unless I'm missing something?

Works without a nexthop using the onlink flag. But I forgot this is not merged
yet, so nevermind (https://github.com/FRRouting/frr/pull/19429).

> >>> the ip for the nic is taken from the node config, so leaving the nic ip
> >>> fields empty doesn't necesarrily mean do p2p, p2p still needs a
> >>> next-htop target and for v4 this has to be specified. we just take the
> >>> ip from the node, which is fine. but leaving the ip field empty does not
> >>> imply do `p2p`, just: use the node-ip for either(p2p or broadcast).
> >>>
> >>> for v6 we could actually never copy the node-ip to the nic(we do that
> >>> now always), this would allow configuring unnumbered v6 broadcast or
> >>> p2p, which are both possible with v6. 
> >>>
> >>> does isisd support v4-via-v6 routes? if it does we could just do v6
> >>> unnumbered in case no nic ip is specified and skip configuring the nic
> >>> ourself..
> >>
> >> Nope.
> > 
> > :/
> 
> Only BGP (and Babel) currently do afaik ;_;
> 
> 
> >>> either way, I still think the nob makes sense here, cause the the
> >>> existance of a specified ip on the nic doesn't really point either way
> >>> (bc or p2p), both need one, and for v6 can work without one
> >>
> >> Hmm yes you're right. We could add a knob like with ospf, where we have "auto"
> >> (the current heuristic to select network-type) and then "point-to-point" and
> >> "broadcast".
> > 
> > sounds good, im just not sure if 'auto' could be interpreted as 'find
> > out if there is multiple in this L2 domain'.. since this would be new,
> > maybe just the two explicit choices might be better, but fine either way
> 
> I feel explicitly setting the mode is better and easier to explain to users as
> well...

On the ospf fabric we have a option, so you can leave the selector empty (with a
background-text "auto"). I'd do the same here. Otherwise the user needs to
manually select point-to-point or broadcast on all interfaces. (We could prefill
it though...)