ppc64el / arm64 - issues with cloud-init setting default route

This bug was originally filed in Launchpad as [LP: #1879933](https://bugs.launchpad.net/cloud-init/+bug/1879933)

Launchpad details

affected_projects = ['maas', 'netplan']

assignee = None

assignee_name = None

date_closed = None

date_created = 2020-05-21T10:52:04.137858+00:00

date_fix_committed = 2020-06-05T14:39:45.303506+00:00

date_fix_released = 2020-06-05T14:39:45.303506+00:00

id = 1879933

importance = undecided

is_complete = False

lp_url = https://bugs.launchpad.net/cloud-init/+bug/1879933

milestone = None

owner = admcleod

owner_name = Andrew McLeod

private = False

status = incomplete

submitter = admcleod

submitter_name = Andrew McLeod

tags = []

duplicates = []

_Launchpad user **Andrew McLeod(admcleod)** wrote on 2020-05-21T10:52:04.137858+00:00_

This is quite possibly a cloud-init bug.

MAAS version: 2.6.2 (7841-ga10625be3-0ubuntu1~18.04.1)

This problem manifests whether to machine is deployed with juju or manually via the MAAS ui.

This problem is intermittent and I have only seen it affecting arm64 and ppc64el machines (out of 29 machines in total) - all of these machines have 2 interfaces connected to the same fabric in the same subnet - one is set to unassigned to be used as a bridge port / data port for openstack deployments, the other is set to auto assign.

This problem occurs with bionic, eoan and focal deployments.

I have recommissioned the affected machines numerous times, including attempts to update firmware.

Symptoms: when the machine comes up after it is deployed there is no default gateway, e.g.

ubuntu@node-mawhile:/var/log$ ip route
10.245.168.0/21 dev enP5p9s0f0 proto kernel scope link src 10.245.168.63

The rsyslog on the MAAS server shows that the machine is being configured correctly:

https://pastebin.ubuntu.com/p/ZZzQ4q2ZCT/

But the cloud-init log on the machine does not have a default gateway:

https://pastebin.ubuntu.com/p/cCJbF7zhtK/

Additional info:

Something I have observed is that the machines where this problem occurs seem to sometimes have the 'unassigned' interface as the PXE interface, and sometimes the auto-assigned interface. I've tried to force this but the PXE interface moves around by itself.