MTU is not set on vlan interface

[Benjamin Allot]

Hello,

Upon changing the following configuration, adding the MTU lines, I ended up with the VLAN interfaces still in MTU 1500 after a `netplan generate && netplan apply`

# lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description: Ubuntu 20.04.1 LTS
Release: 20.04
Codename: focal

# uname -a
Linux il3-fw2 5.4.0-52-generic #57-Ubuntu SMP Thu Oct 15 10:57:00 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux

# dpkg -l netplan.io
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name Version Architecture Description
+++-==============-======================-============-===========================================================
ii netplan.io 0.100-0ubuntu4~20.04.2 amd64 YAML network configuration abstraction for various backends

==========

# This is the network config written by 'subiquity'
network:
  ethernets:
    ens10f0: {}
    ens10f1: {}
    ens10f2: {}
    ens10f3: {}
    ens1f0: {}
    ens1f1: {}
  bonds:
    bond-oam:
      interfaces: [ens10f0, ens10f1]
      parameters:
        mode: 802.3ad
        transmit-hash-policy: layer3+4
        mii-monitor-interval: 100
        lacp-rate: fast
      mtu: 9000

    bond-manlan:
      interfaces: [ens10f2, ens10f3]
      parameters:
        mode: 802.3ad
        transmit-hash-policy: layer3+4
        mii-monitor-interval: 100
        lacp-rate: fast

    bond-core:
      interfaces: [ens1f0, ens1f1]
      parameters:
        mode: 802.3ad
        transmit-hash-policy: layer3+4
        mii-monitor-interval: 100
        lacp-rate: fast
      mtu: 9000

  vlans:
    vlan3001:
      id: 3001
      link: bond-manlan
      addresses: [ "10.130.1.3/24" ]
    vlan3002:
      id: 3002
      link: bond-manlan
      addresses: [ "10.130.2.3/24" ]
    vlan3003:
      id: 3003
      link: bond-manlan
      addresses: [ "10.130.3.3/24" ]

    vlan3054:
      id: 3054
      link: bond-core
      addresses: [ "10.130.56.3/24" ]
      mtu: 9000

    vlan3065:
      id: 3065
      link: bond-oam
      addresses: [ "10.130.63.3/24" ]
      mtu: 9000

  version: 2
==========

And here is the result after the `netplan apply`

===============
# netstat -in
Kernel Interface table
Iface MTU RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg
bond-cor 9000 19448696 0 0 0 6858767 0 1 0 BMmRU
bond-man 1500 7981199 0 1 0 1755800 0 2 0 BMmRU
bond-oam 9000 6236232 0 0 0 4488877 0 1 0 BMmU
ens10f0 9000 3961100 0 0 0 2271891 0 0 0 BMsU
ens10f1 9000 2288169 0 0 0 2218120 0 0 0 BMsU
ens10f2 1500 2506083 0 0 0 1226075 0 0 0 BMsRU
ens10f3 1500 5475116 0 0 0 529725 0 0 0 BMsRU
ens1f0 9000 9513761 0 0 0 3970465 0 0 0 BMsRU
ens1f1 9000 9934935 0 0 0 2888302 0 0 0 BMsRU
lo 65536 21336 0 0 0 21336 0 0 0 LRU
vlan3001 1500 1952436 0 0 0 1455753 0 0 0 BMRU
vlan3002 1500 251466 0 0 0 2617 0 0 0 BMRU
vlan3003 1500 299241 0 0 0 2432 0 0 0 BMRU
vlan3051 1500 5026690 0 0 0 17535 0 0 0 BMRU
vlan3054 1500 90889 0 0 0 1812 0 0 0 BMRU
vlan3060 1500 9759629 0 0 0 1646313 0 0 0 BMRU
vlan3065 1500 913398 0 0 0 229350 0 0 0 BMUthe netplan apply
===============

However, we can see that the `netplan generate` worked:
=============
# cat /run/systemd/network/10-netplan-vlan30{54,65}.net{dev,work} | grep -i MTU
MTUBytes=9000
MTUBytes=9000
MTUBytes=9000
MTUBytes=9000
==============

[Dimitri John Ledkov (xnox)]

I wonder if you need more mtu settings on:
* ens10f2
* ens10f3
* bond-manlan

I don't think that MTU is allowed to be higher on a vlan, than on bond-man, than on physical interfaces. Why did you not set mtu: 9000 on bond-manlan?

however that does not explain how come the vlans on top of bond-core / bond-oam are not 9000.

[Dimitri John Ledkov (xnox)]

this could be udevd/networkd bug which both fiddle with mtu.

[Benjamin Allot (ballot)]

Hello,

I don't need it on bond-manlan for sure (MTU 1500 on this one and its VLANs), but I read in several place that setting MTU on the bond would also set it on interface members.
And that indeed worked correctly.

As I said, I think the issue is definitely more in the udevd/networkd part (remind me of the azure issue seen in https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1902960) because the /run/systemd/network files were correct after the netplan apply.

[Dan Streetman (ddstreet)]

I can't reproduce it, on a VM with similar config the vlan does get the right mtu. can you provide your systemd-networkd journal log? Is there a specific environment (e.g. specific cloud) this fails in?

[Lukas Märdian (slyon)]

This looks kind of similar to LP: #1888726

[Benjamin Allot (ballot)]

Sadly, the journalctl logs don't go back that far now.

It failed on a baremetal server, not a cloud.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Dan Streetman (ddstreet)]

please reopen if this is still an issue