3 releases

0.0.3 Dec 25, 2020
0.0.2 Dec 20, 2020
0.0.1 Dec 15, 2020

#1455 in Algorithms

26 downloads per month

MIT license

60KB
1K SLoC

Presentation

This crate implements a gossip protocol that is configurable and generic with respect to the application. All aspects of the gossip protocol can be customized (push/pull, gossip period,...), and any data that can be represented in a binary format can broadcast to the network.

The overlay network between the peers is created using Gossip-based Peer Sampling [1]. The configuration for peer sampling can also be customized.

Gossip algorithms

Gossip algorithms enable the propagation of information in a network. It is achieved by nodes exchanging their knowledge. At each round, a node selects a peer at random in the network and sends him the state of his knowledge. The peer responds with its own view, and each peer is able to request the parts of the view that are missing on his side.

Gossip-based Peer Sampling

In large distributed systems that use gossip protocols to broadcast information to the network, peers should be selected at random in the network. In theory, this requires a knowledge of all the participating nodes. Because peers are constantly coming and leaving, this can be impractical. Gossip-based Peer Sampling is an algorithm by Jelasity[1] et al. that solves the random peer selection problem by building a local view of the network that simulates a random view. The algorithm consists of rounds of push and/or pull when peers exchange their views of the network. During each round a node selects a peer at random and either sends him (i.e. push) a selection of peers inside its view (if push is enabled) or an empty view to trigger a pull (if push is disabled). The selected node will process the view received (possibly empty), and respond with its own view if pull is enabled.

API

The gossiping functionalities are provided by the GossipService struct:

  • start starts the gossip protocol on the node
  • submit broadcasts an update to the network
  • shutdown terminates the gossip protocol on the node

Initialization

To join an existing network, a new node must connect to at least one existing peer to learn about other peers. This is done by providing the start method with a closure that returns a list of Peer. The peer(s) can be hardcoded or retrieved via any other means inside the closure.

If no Peer is returned by the closure, the node will wait for connections from other peers.

Receiving updates from the network

Updates broadcast by other peers must be delivered to the application layer. To this end, the start method also requires a struct implementing the UpdateHandler trait to handle Update messages received from other peers.

Example

Implementing a simple handler for text messages

pub struct MyUpdateHandler;

impl UpdateHandler for MyUpdateHandler {
    fn on_update(&self, update: Update) {
        let _string_message = String::from_utf8(update.content().to_vec()).unwrap();
        // do something with the message...
    }
}

Starting a gossip node

fn main() {
    // local machine IP and port for listening
    let my_address = "127.0.0.1:9000";
    
    // existing peer(s) in the network
    let existing_peers = || Some(vec![ Peer::new("127.0.0.1:9001".to_owned()) ]);
    
    // create and start the service
    let mut gossip_service = GossipService::new_with_defaults(address.parse().unwrap());
    gossip_service.start(Box::new(existing_peers), Box::new(MyUpdateHandler))?;
    
    // submit a message
    gossip_service.submit("Some random message".as_bytes().to_vec())?;

    // shutdown the gossip protocol on exit
    //gossip_service.shutdown();
}

[1]: M. Jelasity, S. Voulgaris, R. Guerraoui, A.-M. Kermarrec, M. van Steen, Gossip-based Peer Sampling, 2007

Dependencies

~4MB
~92K SLoC