Trying to setup a highly available Zigbee network

[comk22]

I posted this in the home assistant forum but not getting much feedback. https://community.home-assistant.io/t/trying-to-setup-a-highly-available-zigbee-network/830724

I have a partly solution but it's far from ideal. Please see at the end of the post.

The details:

title says most of it. This is my current setup:

• 2x Sonoff Dongle Plus same firmware each connected to
• 2x Rpi with a virtual IP and keepalived setup
• Zigbee2Mqtt on my Hassio instance on an external server is accessing the dongle via the virtual IP and TCP and not through physical USB.

Setup with 1 dongle and rpi worked well before.

Since there are important sensors running on my network I wanted to have a fallback dongle. So in case one dongle goes down the other one on the 2nd RPI will take over. This is all organized with a virtual IP and the keepalived package which monitors basically a ping to the other RPI.

I flashed the second dongle with the same firmware and ieee as described in the docs of z2m with cc2538-bsl.

So far so good. The Rpi config and virtual IP work flawlessly and the backup dongle is detected successfully by z2m when I poweroff my primary RPI. BUT I just can’t get it to work properly. The backup dongle won’t react to switches and buttons properly if at all. Some lights have a long delay other devices won’t react in any way.

This is the most common error I’ll get:

2025-01-17 16:43:03] error: 	z2m: Publish ‘set’ ‘state’ to ‘Arbeitszimmer Luftreiniger’ failed: ‘Error: ZCL command 0x70ac08fffe9484cb/1 genOnOff.off({}, {timeout:10000,disableResponse:false,disableRecovery:false,disableDefaultResponse:false,direction:0,reservedBits:0,writeUndiv:false}) failed (Timeout - 8714 - 1 - 15 - 6 - 11 after 10000ms)’
2025-01-17 16:43:03] error: 	z2m: Publish ‘set’ ‘state’ to 'Schlafzimmer Luftreiniger ’ failed: ‘Error: ZCL command 0xf4b3b1fffea92a71/1 genOnOff.on({}, {timeout:10000,disableResponse:false,disableRecovery:false,disableDefaultResponse:false,direction:0,reservedBits:0,writeUndiv:false}) failed (Data request failed with error: ‘No network route’ (205))

The rest of the logs (activated debug logging) look normal. Motion sensors and power plugs with current sensors will show up normally.

I also tried this only on my primary RPI to rule out a config mistake with the virtual IP/keepalived setup. So I just unplugged the primary dongle and connected the backup dongle, started Z2M but no joy.

There is a quirk which will make the secondary dongle work though. As described in the z2m docs sometimes devices must be power cycled after migrating a dongle. This worked for me only by power cycling one of my many devices. But this is very inconsistent. Sometimes it works sometimes not. But I don't believe in randomness and I'd like to understand this better.

Why does the network need a power cycle? How can I replicate this behaviour to be consistent so that my secondary dongle will work as expected?