UDP auto-discovery for peers on the same machine

Question

I'm looking at ZeroMQ Realtime Exchange Protocol (ZRE) as inspiration for building an auto-discovery of peers in a distributed application.

I've built a simple prototype application using UDP in Python following this model. It seems it has the (obivious, in retrospect) limitation that it only works for detecting peers if all peers are on other machines. This due to the socket bind operation on the discovery port.

Reading up on SO_REUSEADDR and SO_REUSEPORT tells me that I can't exactly do this with the UDP broadcast scheme as described in ZRE.

If you needed to build an auto-discovery mechanism for distributed applications such that multiple application instances (possibly with different versioN) can run on the same machine, how would you build it?

Answer 1

You should be able to bind each server instance to a different address. The entire subnet 127.0.0.0/8 should resolve to your localhost, so you can set up - for example - one service listening on 127.0.0.1, another listening on 127.0.0.2, etc. Anything from 127.0.0.1 to 127.255.255.254.

# works as expected
nc -l 127.0.0.100 3000 &
nc -l 127.0.0.101 3000 &

# shows error "nc: Address already in use"
nc -l 127.0.0.1 3000 &
nc -l 127.0.0.1 3000 &