field link-local not found in type netplan.Ethernet

[Arif Ali]

juju version: 2.8.9
MAAS verwsion: 2.9.2
series: bionic

# Problem

When creating an LXD container by running the following command

juju add-machine lxd:0 --constraints "spaces=oam"

I get an error below

machine 0/lxd/1 failed to start: failed to bridge devices: bridge activation error: yaml: unmarshal errors:
  line 5: field link-local not found in type netplan.Ethernet
  line 7: field link-local not found in type netplan.Ethernet
  line 9: field link-local not found in type netplan.Ethernet
  line 11: field link-local not found in type netplan.Ethernet
  line 13: field link-local not found in type netplan.Ethernet
  line 15: field link-local not found in type netplan.Ethernet
  line 17: field link-local not found in type netplan.Ethernet

# What I expect

I expect that the LXD container to be created

# What I did to get to this point

I have been trying to get around the issue for LP 1831892. When you have a machine deployed with bridges already deployed on the spaces, and then create LXD container as above, the container is created, and there is no problems.

# Steps to re-produce

1. Add a new machine using "juju add-machine"
2. juju ssh to the machine, and update all the interfaces so that "link-local: []", similar to [1]
3. Then run "netplan apply"
4. Now, run the command below, and we will get the error as shown above

juju add-machine lxd:0 --constraints "spaces=oam"

# Logs

Enabled TRACE logging using the command below, and the relevant logs from the machine are in [2]

juju model-config logging-config="<root>=TRACE"

# Workaround

1. Remove the link-local configuration from netplan configuration (without re-applying, as the link-local configuration would gave been applying by netplan originally)
2. Add the lxd container as per usual (This will now create the bridge)
3. Add the netplan configuration again (without-reapplying)

Now, all LXD containers will work as expected

[1] https://paste.ubuntu.com/p/tB35m3BwYp/
[2] https://paste.ubuntu.com/p/87JjDXFbn5/

Related Bug: #1831892

[Achilleas Anagnostopoulos (achilleasa)]

PR https://github.com/juju/juju/pull/12772 adds `link-local` to the supported set of fields when unmarshalling netplan configurations. This should allow the above 'juju add-machine' invocation to work as expeected.

[Arif Ali (arif-ali)]

Hi @achilleasa,

Thanks for this, and have gotten around to test this today

I have, unfortunatley encountered a problem. When I have a seperate file in /etc/netplan with the extra configuration for netplan with link-local, adding the lxd machine fails, and the hypervisor loses all networking

Below is the log from the hypervisor (I added a password, and logged in manually after the fact, to get the details)
https://paste.ubuntu.com/p/kCs9CXdWWw/

Below is my juju model-config cloudinit-userdata
https://paste.ubuntu.com/p/prcc586HfY/

As you can see, I am adding an extra file called /etc/netplan/99-post-juju.yaml, and adding the link-local there.

But, when the lxd container is created, and the bridging re-done by juju, I get the following in /etc/netplan/99-juju.yaml

https://paste.ubuntu.com/p/SwfRh7K3h2/

Which is unexpected. I would expect it to combine the configs together.

Further on, when I add the link-local: [] to 50-cloud-init.yaml, after I have deployed it. juju does not respect the updated configuration, and the new configuration created as 99-juju.yaml will not have link-local, as shown below

root@as2-maas-node-06:/etc/netplan# grep link 99-juju.yaml
root@as2-maas-node-06:/etc/netplan# grep link 50-cloud-init.yaml.bak.1617018956
            link-local: []
            link-local: []
            link-local: []
            link-local: []
            link-local: []
            link-local: []
            link-local: []

Any thoughts on this would be appreciated

Happy to test this, as this is a local lab environment where I have deployed

[Achilleas Anagnostopoulos (achilleasa)]

AFAICT juju does not currently support merging of multiple netplan config files (https://github.com/juju/juju/blob/2.9/network/netplan/netplan.go#L360-L363) and instead replaces any entries it has read so far (last read wins).

There is a relevant old bug about this https://bugs.launchpad.net/juju/+bug/1701429 but it is marked as wont-fix. I will ask around to see why this was never implemented.

[Achilleas Anagnostopoulos (achilleasa)]

I have re-opened https://bugs.launchpad.net/juju/+bug/1701429 to support merging of netplan configurations.

[Arif Ali (arif-ali)]

Thanks for your effort on this :)

that will help with the first issue I had.

The second issue was, although juju did not complain about "link-local: []" in the netplan from /etc/netplan/50-cloud-init.yaml (with my manual changes), it did not carry over the config to 99-juju.yaml.

[Achilleas Anagnostopoulos (achilleasa)]

@arif-ali I proposed a PR for LP1701429 that should take care of these issues.