gossip

topic-based publish/subscribe over epidemic broadcast trees

built on two papers: HyParView for hybrid partial view peer sampling, and PlumTree for efficient broadcast tree construction on top of the random peer graph

topics

a topic is a 32-byte identifier defining a gossip swarm. subscribe to a topic to receive all messages broadcast to it. topics partition the network into overlapping interest groups

interface

subscribing returns a (Sender, Receiver) pair. Sender broadcasts messages to the topic. Receiver streams incoming messages as events: Received (message + sender identity), Peer (new peer discovered), Subscribed (joined topic confirmation)

propagation

builds a broadcast tree overlay on top of the random peer graph. eager push along tree edges, lazy push on the remaining links — combines the reliability of flooding with the efficiency of tree routing. repairs itself when peers leave or join

transport

uses the ALPN identifier "gossip" over QUIC bidirectional streams routed through radio/router. connections established via radio/endpoint

role in cyber

gossip is how neurons learn about new cyberlinks in real time. when a neuron creates a link, it broadcasts to the relevant topic. subscribers update their local view of the cybergraph without polling. also serves as the notification layer for radio/docs synchronization — when a replica entry changes, gossip carries the signal

crate: iroh-gossip