Our customer tends to use OSPF between PE-CE on EVPN. Using here is:

- EVPN All-Active multihoming

- L3-interfaces irb, which is located in VRF and is running OSPF.

Issue we noticed:

- Both CE and 2 PEs, cannot fully discover all other nodes.

- In some cases where CE can detect one PE (state 2Way achieved), it can transition into a FULL state or Exstart state base on which the DD packets are sent.

Topology:

PTX <-EVPN> MX & both connected to same CE via LAG

MX-CE (with LAG to PTX down)
+++++++++++++++
labroot@jtac-mx480-r2044-re0> show ospf neighbor logical-system CE-1093    
Address          Interface              State           ID               Pri  Dead
10.12.249.10     xe-4/0/3.1094          Full            10.12.249.10     128    37
10.12.249.11     xe-4/0/3.1094          Init            10.12.249.11     200    32 <<<PTX-PE

labroot@jtac-mx480-r2044-re0> show interfaces descriptions 
Interface       Admin Link Description
xe-4/0/0        down  down et-0/0/0-jtac-ptx10001-36mr-k-r2016
xe-4/0/1        up    up   xe-4/0/0:0-jtac-mx480-r2043-mx2
PTX-PE (with LAG to PTX down)
labroot@jtac-ptx10001-36mr-k-r2016> show ospf statistics instance POC_EVPN_VRF_OOB    

Packet type             Total                  Last 5 seconds
                   Sent      Received        Sent      Received
   Hello              4             0           0             0 <<<No Hellos received
     DbD              0             0           0             0
   LSReq              0             0           0             0
LSUpdate              0             0           0             0
   LSAck              0             0           0             0

The PTX can have from peering over the direct link with CE, as it do not need to pass evpn tunnel.

But when the direct link is down, the flooded hello from MX over even tunnel is no accepted.

MX currently support this & the PTX platform do not have support for same.

There is no plan for PTX to include this as of now.

This is why MX PE peering comes up when MX PE to CE link is down & hellos are via PTX.

Below is the extract from internal RLI page, which mention this point.

++++++++

The following features are not supported:

1. Any protocol support (BGP/OSPF/ISIS) between CE and remote PE over EVPN MPLS tunnel is not supported. This would apply to cases when CE is connected to MH PEs as the protocol packets may be hashed to one MH PE and would need to be transported to the other active MH PE over EVPN tunnel – hence CE – PE protocols may not always work between CE and MH PEs in this release.