Nebula interface ignored by Avahi

[hmoffatt]

I wanted to play around with mDNS over the overlay network, but when I start avahi-daemon it always ignores the nebula interface. The logs show it is listening on all of the others, lo, ethernet, podman0, Docker veth etc but never the nebula.

The only things I note from ip link are

• all the others have type link/loopback or link/ether, while the nebula interface has link/none
• all the others have state UP while the nebula interface has state UNKNOWN

I didn't check the Avahi source to see what it's filtering on when discovering interfaces.

Does anyone else have Avahi working over Nebula?
In production I'd rather use systemd-resolved, but that's phase 2.

[jasikpark]

mDNS relies on IP multicast, which Nebula does not support. Being a peer-to-peer protocol, there is no way to send a single packet that is delivered to all machines on the network.

[hmoffatt]

Ok, thanks. What is the function of the config setting tun.drop_multicast then?

[nbrownus]

@hmoffatt that setting exists to end packet process of multicast packets early and avoid wasting resources. If tun.drop_multicast is false then nebula will try to handle the packet, there won't be a tunnel for the address so it will ask your lighthouse and the lighthouse will also not know of a host at that address. The end result is a few noisy log lines (handshake sent, handshake timed out) and wasted resourced on both the local host and the lighthouse(s).

Setting tun.drop_multicast to true avoids this situation.