raft

This is an implementation of the Raft distributed consensus protocol. It's heavily influenced by benbjohnson's implementation. It focuses on providing a clean and usable API, and well-structured internals.

Usage

A node in a Raft network is represented by a Server structure. In a typical application, nodes will create a Server, and expose it to other nodes using a Peer interface.

Servers are only useful when they can communicate with other servers. This library includes a HTTP Transport (ingress) and HTTP Peer (egress) which combine to allow communication via REST-ish endpoints. For now, it's the simplest way to embed a Raft server in your application. See this complete example.

Several other transports are coming; see TODO, below.

Adding and removing nodes

The Raft protocol has no affordance for node discovery or "join/leave" semantics. Rather, the protocol assumes an ideal network configuration that's known a priori to nodes in the network, and describes a mechanism (called joint-consensus) to safely replicate that configuration.

My implementation of joint-consensus abides those fundamental assumptions. Nodes may be added or removed dynamically by requesting a SetConfiguration that describes a complete network topology.

TODO

• ~~Leader election~~ done
• ~~Log replication~~ done
• ~~Basic unit tests~~ done
• ~~HTTP transport~~ done
• net/rpc transport
• Other transports?
• ~~Configuration changes (joint-consensus mode)~~ done
• Log compaction
• Robust demo application ☜ in progress
• Complex unit tests (one per scenario described in the paper)