Hello @Network-learner !
I'll try to simplify this a bit from my understanding of how it works .
The "magic" happens when you actually map a VLAN to a L2VNI or a VRF to a L3VNI - depending on your scenario.
Let's take VLAN 10 for example - traffic comes into a port and arrives tagged or the switches places the tag of VLAN 10 on that packet . After that the switch strips the VLAN tag (not taking into account Q-in-VNI or other complex scenarios here) and then places the VXLAN header over it . After this the traffic is processed based on the information of "NVE" interface - it searches for all the VTEP peers for which it exchanges information for that specific VNI . After one or more peers are detected - the traffic is encapsulated to these using the source-interface of the NVE interface and sent via multicast or unicast - depending on if the traffic is either BUM traffic or plain unicast traffic. Note that these VTEP adjacencies are usually established beforehand .
Comparing this to the GRE tunnel interface might not be the best idea - usually over a GRE interface you do routing - and it makes sens that you see the traffic "outgoing" via the GRE tunnel interface in the routing table . It's not the same for VXLAN - you're talking about Layer 2 and Layer 3 traffic both in that case. Long story short - you map your VLAN to you VXLAN and then send the traffic out using the source IP address of the NVE interface towards the peer switch you have a VTEP adjacency with.
Hope that clarifies some things for you.
Best regards!
Horia
