k3s does not start in airgapped set up

[PidgeyBE]

Hi

When deploying k3s via EIB, k3s will not start if there is no ethernet cable plugged in into the device.
This results in:

[FAILED] Failed to start Network Manager Wait Online.
[FAILED] Failed to start Lightweight Kubernetes.

When inspecting journalctl -u k3s I see
level=fatal msg="no default routes found in "/proc/net/route" or "/proc/net/ipv6_route"

Am I doing something wrong or is EIB not made to be used in fully airgapped situations?
Even if I install the same ISO on a device with ethernet cable plugged in, I get the same issues on first boot. After rebooting, k3s does start.

BR, Pj

[atanasdinov]

Hello! EIB is definitely meant to be used in airgapped environments. It will, however, not assume there's a missing network. One of the main requirements for both k3s and RKE2 is to have a default route set up -- please see https://docs.k3s.io/installation/airgap#default-network-route.

There are numerous ways of configuring the network with EIB, please check our official docs to figure out what's the proper approach for your use case.

[jdob]

@PidgeyBE If you're comfortable with Atanas' answer, can you please close out the issue?

[PidgeyBE]

Definitely!

I've been using the information shared by Atanas to try to come to a working setup, but unfortunately I haven't succeeded yet.
A sample configuration of how to exactly configure EIB to have a dummy network interface would be very helpful.

[atanasdinov]

You can follow the custom configuration e.g. something along the lines of:

mkdir -p $CONFIG_DIR/network

cat << EOF > $CONFIG_DIR/network/configure-network.sh
#!/bin/bash
set -eux

ip link add dummy0 type dummy
ip link set dummy0 up
ip addr add 203.0.113.254/31 dev dummy0
ip route add default via 203.0.113.255 dev dummy0 metric 1000
EOF

chmod a+x $CONFIG_DIR/network/configure-network.sh

This will use the "custom network script" functionality of EIB which in turn executes the example from the k3s docs about a dummy route. I haven't tested it but I'd expect this to work.

[PidgeyBE]

Yeah, I've tried something similar before, but it didnt work.
Reproduced the issue with your example:

[atanasdinov]

Can you show the output of these two:

1. journalctl -u combustion-prepare
2. journalctl -u combustion

These are first things to check when there's a failure in the boot process.

[PidgeyBE]

Thanks, good to know!
Seems like it doesn't know the device type dummy...

[PidgeyBE]

I fixed this by creating network/_all.yaml with:

interfaces:
  - name: eth0
    type: ethernet
    state: up
    ipv4:
      enabled: true
      dhcp: true
  - name: dummy0
    type: dummy
    state: up
    ipv4:
      enabled: true
      address:
        - ip: 203.0.113.254
          prefix-length: 31
routes:
  config:
    - destination: 0.0.0.0/0
      next-hop-address: 203.0.113.255
      next-hop-interface: dummy0
      metric: 2000

That works.

Although, in case a network cable is connected, the eth0 does not get an IP address on first boot. On second boot the interface works...
So that's still a remaining issue