On QFX5k, VXLAN ERB host route for some remote mac-ip pairs is pointing to discard next-hop after Junos upgrade to 21.4R3-S4.

From shell, we check using this command and get below output:

i.e for destination 10.0.0.1 with next-hop 16137

% cprod -A fpc0 -c "show route ip" | grep 10.0.0.1

10.0.0.1 Discard 16137 RT-ifl 668

At the same time, in UKERN_GBL.log log file, we see the following traces for the affected destinations: 

[Thu Aug 3 12:59:51.937] [9255917] brcm_nh_ucast_initialize:1829 NH-HALP : BRCM_NH-NH 16137 type 0 proto 0 bnbe_nh_idx 0

[Thu Aug 3 12:59:51.937] [9255918] brcm_nh_create_l2params:157 NH-HALP : BRCM_NH-nh_type:Discard, nh_params_type:Discard, nh_id:16137, flabel:0, num_tags 0

[Thu Aug 3 12:59:51.937] [9255919] brcm_nh_ucast_install:1370 NH-HALP : BRCM_NH-nh:16137 type:0(Discard) proto:0(Unknown)

[Thu Aug 3 12:59:51.937] [9255920] brcm_nh_type_supported:380 NH-HALP : BRCM_NH-type:Discard, Id:16137 Transient Discard NH allowed

[Thu Aug 3 12:59:51.937] [9255921] brcm_nh_set_shim_data:561 NH-HALP : BRCM_NH-NH 16137 type 0 proto 0 ifd .local.

[Thu Aug 3 12:59:51.937] [9255922] brcm_nh_ucast_install:1432 NH-HALP : BRCM_NH-type:Discard, Id:16137 Transient Discard NH allowed

Issue resolved through PR1749759.

Restoration steps

Clear problematic Mac-ip entries either of these way:

- manual clear from all MH leaf where learned locally.

- apply policy to reject and remove.

- clear overlay BGP so all entries deleted and re added

2023-09-07: initial publication

2024-10-24: added public PR link