I have Cisco Nexus tor switch connected with multiple servers on 25G nic. When I am pinging from server1 to server2 I am seeing higher latency in ping. After tcpdump found server replying ping reply with tor switch gateway mac address. In short MAC getting changed which redirecting packet via VRF1.
In tcpdump i found
ping 1.1.1.3 from 1.1.1.2 [ src: b:b:b:b dst: c:c:c:c ]
when server2 reply back it use [ src: c:c:c:c dst: a:a:a:a ]
Technically it should use server1 mac address in icmp reply because they are in local LAN. This is EVPN fabric leaf switch. I have native + tagged vlan configured on switch.
#EDIT
interface Vlan10 no shutdown mtu 9216 vrf member MGMT ip address 10.10.10.1/24 fabric forwarding mode anycast-gateway interface Vlan11 no shutdown mtu 9216 vrf member L3GW ip address 1.1.1.1/24 fabric forwarding mode anycast-gateway interface Ethernet1/1 description server1 switchport switchport mode trunk switchport trunk native vlan 10 spanning-tree port type edge trunk spanning-tree bpduguard enable mtu 9216 interface Ethernet1/2 description server2 switchport switchport mode trunk switchport trunk native vlan 10 spanning-tree port type edge trunk spanning-tree bpduguard enable mtu 9216 I have notice one thing on server2 in ip neigh output where entry is STALE. We have arp suppression enabled on cisco evpn fabric.
root@server2:~# ip neigh show | grep b:b:b:b 1.1.1.2 dev vlan11 lladdr b:b:b:b STALE 10.10.10.2 dev ens2f0 lladdr b:b:b:b REACHABLE 
{}preformat option).