Hi, > On 23/02/2020, at 11:38 PM, Roger Wiklund <[email protected]> wrote: > > Hi Nathan > > You're not missing anything. EVPN Overlay IRB VGA selection does NOT take > underlay metrics into consideration yet (it's on the roadmap) > What you need to do is to filter out the remote DC IRB VGAs on each leaf > switch using a policy. > > Leaf switches: > set policy-options policy-statement OVERLAY-IN term reject-remote-gw from > family evpn > set policy-options policy-statement OVERLAY-IN term reject-remote-gw from > next-hop [ 100.0.0.13 100.0.0.14] <--Loopback VTEP IPs of remote DC you want > to reject > set policy-options policy-statement OVERLAY-IN term reject-remote-gw from > nlri-route-type [ 1 2 ] > set policy-options policy-statement OVERLAY-IN term reject-remote-gw then > reject > set policy-options policy-statement OVERLAY-IN term accept-all then accept > > Add this OVERLAY-IN as an import policy for your BGP Overlay group. > > *Note, this policy filters out ALL route type1 and route type2 routes, so not > only the VGA MAC. After 19.4R1 you can enhance this policy to only filter out > the VGA MAC from the remote DC. > This is useful if you have L2 stuff connected directly to Spine that you want > to stretch to both DC1 and DC2. With the above policy those L2 VNIs will be > rejected. > > https://www.juniper.net/documentation/en_US/junos/topics/concept/evpn-routing-policies.html > > <https://www.juniper.net/documentation/en_US/junos/topics/concept/evpn-routing-policies.html> > > Regards > Roger
Hah! I literally arrived at this solution just now myself, and am checking email before calling it a day and going to approach it again tomorrow. Just been through a “let’s try ebgp everywhere” loop, and a few other “let’s try this a whole different way” attempts. Thanks for the (unintentional) sanity check! Our spines run as RRs for the local leaf switches, the spines have a full mesh - could I put such a policy on the spine, or, does it need to be on leaves? It would be *fantastic* for underlay metrics to be part of route selection, it seems.. well.. kind of crazy that it’s not in there already. For networks with a folded CLOS topology this seems fine.. until a leaf-spine link failure which could result in some quite poor paths, I think. For different topologies, i.e. ones with different path lengths, this is regularly going to result in poor selected paths. -- Nathan Ward _______________________________________________ juniper-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/juniper-nsp
