This article describes the following scenario experienced by a customer: after upgrading QFX5120 switches from 22.4R3 to 23.4R2-S4, a subset of endpoints in one VNI may intermittently fail to reach endpoints in another VNI (inter-VN routing). The issue can be limited to hosts connected to a specific leaf/ToR and may clear after the fabric fully reconverges.

> Only some hosts in a VN cannot reach a host/subnet in another VN, while other hosts in the same VN work normally.

> Affected hosts can still:

• reach the default gateway
• show expected ARP/MAC learning after bouncing links

> Issue begins around/after an upgrade maintenance window and may self-resolve later.

> In some cases, control-plane looks correct, but forwarding is not (route present in RIB but not properly installed in FIB/PFE).

A transient post-upgrade reconvergence/programming condition on the ingress leaf where the affected endpoints are connected (VRF route/next-hop temporarily stale or not fully programmed in hardware).

Verify route vs forwarding (on the ingress leaf)

• RIB (VRF route present/active):

show route table <SRC_VRF>.inet.0 <DST_IP> extensive

• FIB/PFE (forwarding programmed):

show route forwarding-table table <SRC_VRF> destination <DST_IP> extensive

If the route is present in the VRF but forwarding is missing/incorrect, proceed with refresh.

• Force a refresh (least disruptive first)

Soft refresh EVPN per peer: clear bgp neighbor <EVPN_PEER_IP> evpn soft-inbound