subiquity crashes if netplan apply fails

[Frank Heimes]

By accident the (external) network of my s390x system was down, hence I decided during a subiquity installation to choose 'Continue without network', expecting that I'm still able to complete the installation, but subiquity crashed.

Then netplan apply failed, and crashes the UI:

Traceback:
 Traceback (most recent call last):
   File "/snap/subiquity/1570/lib/python3.6/site-packages/subiquity/core.py", line 165, in run
     super().run()
   File "/snap/subiquity/1570/lib/python3.6/site-packages/subiquitycore/core.py", line 680, in run
     self.urwid_loop.run()
   File "/snap/subiquity/1570/usr/lib/python3/dist-packages/urwid/main_loop.py", line 286, in run
     self._run()
   File "/snap/subiquity/1570/usr/lib/python3/dist-packages/urwid/main_loop.py", line 384, in _run
     self.event_loop.run()
   File "/snap/subiquity/1570/usr/lib/python3/dist-packages/urwid/main_loop.py", line 1484, in run
     reraise(*exc_info)
   File "/snap/subiquity/1570/usr/lib/python3/dist-packages/urwid/compat.py", line 58, in reraise
     raise value
   File "/snap/subiquity/1570/usr/lib/python3.6/asyncio/events.py", line 145, in _run
     self._callback(*self._args)
   File "/snap/subiquity/1570/lib/python3.6/site-packages/subiquitycore/async_helpers.py", line 25, in _done
     fut.result()
   File "/snap/subiquity/1570/lib/python3.6/site-packages/subiquitycore/controllers/network.py", line 410, in _apply_config
     await arun_command(['netplan', 'apply'], check=True)
   File "/snap/subiquity/1570/lib/python3.6/site-packages/subiquitycore/utils.py", line 85, in arun_command
     raise subprocess.CalledProcessError(proc.returncode, cmd)
 subprocess.CalledProcessError: Command '['netplan', 'apply']' returned non-zero exit status 1.

[Michael Hudson-Doyle (mwhudson)]

I think this is the requests library misbehaving, it's only supposed to raise subclasses of requests.exceptions.RequestException but clearly sometimes it doesn't. We can make our except clauses broader I guess.

[Dimitri John Ledkov (xnox)]

I guess reproducer is static ip configuration with invalid DNS server

[Michael Hudson-Doyle (mwhudson)]

The tracebacks in the description are not what is crashing the UI, that seems to be because netplan apply is failing. That shouldn't be crashing the UI either!

[Michael Hudson-Doyle (mwhudson)]

Here is the netplan that is causing the failure:

 network:
   ethernets: {}
   version: 2
   vlans:
     encc000.2653:
       id: 2653
       link: encc000
       nameservers:
         addresses:
         - 10.245.236.1

this does indeed look bogus: it has a vlan, but no definition for the physical nic.

[Frank Heimes (fheimes)]

Yes that looks strange, since it does not have the physical base device, but also IP address and gateway (gateway4) are missing.

The /etc/netplan/50-cloud-init.yaml files that I largely saw during my installations looked like this:

# This file is generated from information provided by the datasource. Changes
# to it will not persist across an instance reboot. To disable cloud-init's
# network configuration capabilities, write a file
# /etc/cloud/cloud.cfg.d/99-disable-network-config.cfg with the following:
# network: {config: disabled}
network:
    version: 2
    ethernets:
        encc000.2653:
            addresses:
            - 10.245.236.15/24
            gateway4: 10.245.236.1
            match:
                macaddress: 36:5e:5e:ce:a0:a0
            nameservers:
                addresses:
                - 10.245.236.1
            set-name: encc000.2653

They have at least an ip address and a gateway.

However, a fully correct one should be close to this:

01-netcfg.yaml
network:
    version: 2
    renderer: networkd
    ethernets:
        encc000:
            dhcp4: no
            dhcp6: no
    vlans:
        encc000.2653:
            link: encc000
            id: 2653
            addresses:
            - 10.245.236.15/24
            gateway4: 10.245.236.1
            nameservers:
                addresses:
                - 10.245.236.1

[Dimitri John Ledkov (xnox)]

I wonder if initrd generated wrong .yaml, because i think in vlan case it does not emit "dhcp4: no\n dhcp6: no" lines, but that should be default.

[Michael Hudson-Doyle (mwhudson)]

The point of this bug vs all the other similar ones was that netplan apply crashing should not crash the UI, that's fixed now. Hopefully all the other network woes are being tracked in other bugs.

[Frank Heimes (fheimes)]

With image from Apr 14th I didn't ran into this anymore.
So looks like this is fixed now - thx.