This is my implementation of the Paxos algorithm for uniform consensus detailed in Leslie Lamports paper "Paxos made simple". In this implementation every node is both an acceptor and a proposer.

We assume that the system is crash-recovery and that it is built alongside an eventual leader elector component which provides TRUST events to the processes so they are aware of the leaders identity.

Once the module has been compiled in Iex you must create an upper layer process (which will be sent the DECIDE events from processes within the system once a decision has been reached) and a list of process names must also be defined in an array. ` pids = pids = Enum.map(procs, fn(p) -> Paxos.start(p, procs, upper_layer) end) Processes can be sent trust events manually through Iex to simulate the eventual leader detector:

send(:global.whereis_name(:name_to_receive), {:trust, :name_to_trust})

Proposals can be made by a process by sending them the following in Iex:

send(:global.whereis_name(:name_to_receive), {:trust, :value_to_propose})

Once a process has been trusted as a leader and has a proposal, abortable consensus begins and up to a minority of crashes can be tolerated. Processes which crash can be "recovered" by starting a new process with the exact same name. When a process is initialised it will check disc memory for a persisted version of the state and recover from file if it exists.

When abortable consensus has been successful the upper layer will receive DECIDE events from the processes.

A variety of different test cases are outlined in the tests.ex file.