0

I'm playing with vxlans on ubuntu 18.04 right now and I just found something strange. I have 2 servers with 4 containers on each. Containers are connected to bridges br1 - br4. There are vx1 - vx4 interfaces connected to corresponding bridges. Whole setup looks like this:

             node1                               node2
------- ------- ------- -------     ------- ------- ------- -------     
| c1a | | c2a | | c3a | | c4a |     | c1b | | c2b | | c3b | | c4b |
------- ------- ------- -------     ------- ------- ------- -------
   |       |       |       |           |       |       |       |
------- ------- ------- -------     ------- ------- ------- -------     
| br1 | | br2 | | br3 | | br4 |     | br1 | | br2 | | br3 | | br4 |
------- ------- ------- -------     ------- ------- ------- -------
   |       |       |       |           |       |       |       |
------- ------- ------- -------     ------- ------- ------- -------     
| vx1 | | vx2 | | vx3 | | vx4 |     | vx1 | | vx2 | | vx3 | | vx4 |
------- ------- ------- -------     ------- ------- ------- -------
   |       |       |       |           |       |       |       |
-------------------------------     -------------------------------     
|            eth0             |     |            eth0             |
-------------------------------     -------------------------------
               |                                   |
               \-----------------------------------/

Interfaces on nodes itself are same (at least configuration of bridges and vxlans) and looks like this:

auto vx1
iface vx1 inet manual
    mtu 1450
    pre-up ip link add vx1 type vxlan id 2584 group 239.0.3.20 dstport 8472 port 32768 61000 dev eth0 || true
    up ip link set vx1 up
    down ip link set vx1 down
    post-down ip link del vx1 || true

auto vx2
iface vx2 inet manual
    mtu 1450
    pre-up ip link add vx2 type vxlan id 1428 group 239.0.3.20 dstport 8472 port 32768 61000 dev eth0 || true
    up ip link set vx2 up
    down ip link set vx2 down
    post-down ip link del vx2 || true

auto vx3
iface vx3 inet manual
    mtu 1450
    pre-up ip link add vx3 type vxlan id 2584 group 239.0.3.50 dstport 8472 port 32768 61000 dev eth0 || true
    up ip link set vx3 up
    down ip link set vx3 down
    post-down ip link del vx3 || true

auto vx4
iface vx4 inet manual
    mtu 1450
    pre-up ip link add vx4 type vxlan id 58996 group 239.0.3.14 dstport 8472 port 32768 61000 dev eth0 || true
    up ip link set vx4 up
    down ip link set vx4 down
    post-down ip link del vx4 || true

auto br1
iface br1 inet manual
    bridge_ports vx1
    bridge_stp off
    bridge_fd 0
    bridge_hello 2
    bridge_maxage 12

auto br2
iface br2 inet manual
    bridge_ports vx2
    bridge_stp off
    bridge_fd 0
    bridge_hello 2
    bridge_maxage 12

auto br3
iface br3 inet manual
    bridge_ports vx3
    bridge_stp off
    bridge_fd 0
    bridge_hello 2
    bridge_maxage 12

auto br4
iface br4 inet manual
    bridge_ports vx4
    bridge_stp off
    bridge_fd 0
    bridge_hello 2
    bridge_maxage 12

This setup AFAIK should be working (all vxlan have different pair of group IP and vxlan id) but system cannot setup vx3 because of duplicity of vxlan id (regardless there is different group ip used). Is there any fix for this or there is strict limit vxlan id must be unique even across multiple group ips?

Ondra Sniper Flidr
  • 2,623
  • 11
  • 18

1 Answers1

0

What's with the odd-sized MTUs? Typically VXLAN requires large frames to handle the extra overhead of the VXLAN header - at least 1600, though we always go with the maximum (usually 9000 or so).

Two other comments, I'm not VXLAN-on-Linux expert sorry, but don't you need some form of TEP for each VXLAN host?

Also why VXLAN? It's already been superseded by Geneve?

Chopper3
  • 100,240
  • 9
  • 106
  • 238