Read-your-writes consistency is difficult to achieve in distributed systems in the presence of crashing nodes and network partitions. Consensus algorithms are a way to establish linearizability of operations and allow safe access to your replicated data. This talk will briefly introduce the consistency problems that arise in systems that rely on asynchronous replication and compare a few different solutions to this problem. Of particular note will be availability and operational concerns. 

After the brief introduction to the subject, the Raft consensus algorithm will be discussed enough to understand the rest of the presentation and the Erlang implementation. Once Raft is explained, the approach taken to implement the protocol in Erlang will be described with a focus on the consensus fsm itself and the tradeoffs involved in this implementation. The remainder of the talk will be discussion about verification using the Erlang Quick-Check eqc_statem behaviour.

Talk objectives: By the end of the talk the listeners should understand why consistency is a non-trivial problem in distributed systems and the basics of the raft consensus algorithm. They should also have gotten an introduction to building and testing stateful distributed systems with Erlang Quick Check. The listeners should come away knowing what types of problems property based testing can find, and what it actually did find in rafter. They should also have a grasp of the the strengths and limitations of this method of verification.

Target audience: Distributed Systems Developers and Erlang Developers