Summary

I'm using K3s with Flannel and set --flannel-iface=tailscale0. This works fine until the Tailscale daemon restarts, which causes Flannel's own interface to go MIA.

Environment

$ k3s -v
k3s version v1.32.3+k3s1 (079ffa8d)
go version go1.23.6

Cluster Configuration:

• 1 server, 3 agents
• Using a Tailscale VPN to connect the nodes to each other (--flannel-iface=tailscale0)

What's going wrong

Nodes in my cluster sometimes experience a network connectivity issue that I usually first recognize by being unable to visit a particular service in the browser. Upon inspection, I've find that the flannel.1 interface is missing on at least one of the involved nodes.

I believe this is caused by an intermittent issue with Tailscale such that the tailscale0 interface is recreated (e.g. when it is restarted, either manually or by automatic upgrades), because I reproduce it by running service tailscaled restart).

Steps to reproduce

• Setup a device with Tailscale
• Install a k3s server or agent with --flannel-iface=tailscale0
• ip addr | grep flannel.1 should show have a result
• Run service tailscaled restart
• ip addr | grep flannel.1 should no longer show a result

Expected behaviour

The flannel interface is not lost or at least recovers in due time when the --flannel-iface is available again.

Actual behavior

The flannel interface is gone until k3s restarts.

Additional notes

The issue title is a bit of a guess; I don't actually know if restarting Tailscale causes the tailscale0 interface to be "recreated", but something is happening with it that breaks the flannel interface that depends on it.