Maas controller drops its network config during node enlisting

[hore23]

During enlisting a new node my maas controller drops its vlan interfaces (in maas, not in system):

from regiond.log:

2023-03-14 12:10:03 maasserver.models.signals.interfaces: [info] xx-mgm (vlan) on maas-controller has been deleted; orphaned by eth0 (physical) on maas-controller.
2023-03-14 12:10:03 maasserver.models.signals.interfaces: [info] xx-pxe (vlan) on maas-controller has been deleted; orphaned by eth0 (physical) on maas-controller.

This results in removing subnet entries from zone.maas-internal.

And this results in failed commissioning:

cloud-init[]: request to http://10-11-12-0--24.maas-internal:5248/MAAS/metadata/2012-03-01/ failed. (…) Name or service not known

Controller has one physical network interface with 2 tagged vlans.

The problem occurs in maas 3.2 - 3.4 (snap). (other versions not tested)

A reboot of the controller fixes missing interfaces/vlans in maas.

Controller netplan:

network:
version: 2
renderer: networkd
ethernets:
eth0: {}
vlans:
xx-pxe:
id: 8
link: eth0
addresses: [10.11.12.1/24]
xx-mgm:
id: 9
link: eth0
addresses: [10.11.13.251/24]
routes:
- to: default
via: 10.11.13.254
nameservers:
addresses: [1.1.1.1]

Also tried assigning a dummy ip to physical eth0 without avail.

Why maas thinks my vlans are orphaned? Any ideas?

[hore23]

The root cause seems to be a problem in gathering hardware ressources from controller node, which is maybe architecture related (arm64). The problem also occurs without vlan setup. After fresh install controller shows no network interfaces/vlans. Restarting maas fixes this, but mac address of eth0 is 00:00:00:00:00:00, which is obviously wrong. Trying to enlist a node crashes network config again. regiond.log:

maasserver.region_controller: [info] Reloaded DNS configuration:
* ip 10.11.12.1 disconnected from maas-controller on eth0

There is no physical up/down of eth0.

Controller runs ubuntu 22.04

Any help highly appreciated

[hore23]

So i finally managed to find a quick & dirty workaround.

Problem comes from network card driver “rk_gmac-dwmac”, it reports mac address 00:00:00:00:00:00.

Output from /usr/share/maas/machine-resources/arm64 is:

"network": {
    "cards": [
        {
            "driver": "rk_gmac-dwmac",
            "driver_version": "6.2.0-odroid-arm64",
            "ports": [
                {
                    "id": "eth0",
                    "address": "00:00:00:00:00:00",
                    "port": 0,
                    "protocol": "ethernet",
                    "supported_modes": [
                        "10baseT/Half",
                        "10baseT/Full",
                        "100baseT/Half",
                        "100baseT/Full",
                        "1000baseT/Full"
                    ],
                    "supported_ports": [
                        "twisted pair",
                        "media-independent"
                    ],
                    "port_type": "twisted pair",
                    "transceiver_type": "external",
                    "auto_negotiation": true,
                    "link_detected": true,
                    "link_speed": 1000,
                    "link_duplex": "full"
                }
            ],
            "numa_node": 0
        }
    ],
    "total": 1
},

Further down in section networks the correct mac address is reported (for eth0 as well as vlan interfaces).
By replacing the binary “…/machine-resources/arm64” with a simple wrapper script, which replaces the wrong mac address with the correct one on the fly, maas works again as expected.

So it’s not a bug in maas, but maybe maas could handle this situation better