netplan apply does not remove defunct VLAN interfaces when --state is specificed

[Brian Turek]

"netplan apply" claims that, when invoked using the "--state" parameter, it will remove virtual interfaces when they are not specified in the new config. This functionality does not work for VLAN interfaces. Here's a reproducible:

# cat /etc/netplan/01-vlan.yaml
network:
  version: 2
  renderer: networkd
  ethernets:
    enp0s8: {} # physical ethernet interface
  vlans:
    enp0s8.998:
      id: 998
      link: enp0s8
    enp0s8.999:
      id: 999
      link: enp0s8

# netplan-apply
# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: enp0s8: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP group default qlen 1000
    link/ether 08:00:27:94:6f:b6 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::a00:27ff:fe94:6fb6/64 scope link
       valid_lft forever preferred_lft forever
3: enp0s8.999@enp0s8: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether 08:00:27:94:6f:b6 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::a00:27ff:fe94:6fb6/64 scope link
       valid_lft forever preferred_lft forever
4: enp0s8.998@enp0s8: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether 08:00:27:94:6f:b6 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::a00:27ff:fe94:6fb6/64 scope link
       valid_lft forever preferred_lft forever

# mkdir /tmp/netplan_state && netplan get | tee /tmp/netplan_state/01-old.yaml
network:
  renderer: networkd
  version: 2
  vlans:
    enp0s8.998:
      id: 998
      link: enp0s8
    enp0s8.999:
      id: 999
      link: enp0s8

Now make edits to /etc/netplan/01-vlan.yaml to remove enp0s8.999

# cat /etc/netplan/01-vlan.yaml
network:
  version: 2
  renderer: networkd
  ethernets:
    enp0s8: {} # physical ethernet interface
  vlans:
    enp0s8.998:
      id: 998
      link: enp0s8

# netplan apply --state /tmp/netplan_state/
# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: enp0s8: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP group default qlen 1000
    link/ether 08:00:27:94:6f:b6 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::a00:27ff:fe94:6fb6/64 scope link
       valid_lft forever preferred_lft forever
3: enp0s8.999@enp0s8: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether 08:00:27:94:6f:b6 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::a00:27ff:fe94:6fb6/64 scope link
       valid_lft forever preferred_lft forever
4: enp0s8.998@enp0s8: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether 08:00:27:94:6f:b6 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::a00:27ff:fe94:6fb6/64 scope link
       valid_lft forever preferred_lft forever

I would expect that enp0s8.999 would be removed from the system.

[Brian Turek (brian-turek)]

I did some digging and the lack of removal can either be called a case of unclear documentation or a bug:

On https://github.com/canonical/netplan/blob/a42eb7860c8bbe69b1a2df695c85a59662898362/netplan/cli/commands/apply.py#L204, it passes self.state (the argument that was passed in on the CLI) as the "prefix" to ConfigManager to parse the old config. However, ConfigManager (https://github.com/canonical/netplan/blob/a42eb7860c8bbe69b1a2df695c85a59662898362/netplan/configmanager.py#L154-L156) is coded to look for YAMLs self.prefix / {lib,etc,run} / netplan.

Thus, in my above example, I needed to place my old YAML in /tmp/netplan_state/etc/netplan rather than directyly in /tmp/netplan_state/

Is this desired behavior?

[Lukas Märdian (slyon)]

Yes this is intended behaviour.

The --state parameter passes a "rootdir" that contains a previous state of /{lib,etc,run}/netplan/*.yaml. I guess we should probably improve the documentation about that.