Study Level RAFT Consensus algorithm with erlang

This code is an implementation of the RAFT consensus algorithm written for learning purposes. A single erlang actor process represents a single RAFT node. So, If you want to create a RAFT cluster with 5 nodes, you need to spawn 5 erlang actor processes.

In this code, each RAFT node communicates with other nodes via erlang message passing. This means that no TCP connection is involved.

This RAFT implementation supports this features below.

• RAFT Basic.
  • Log Replication (Append Entries RPC)
  • Leader Election (Request Vote RPC)
• RAFT Advanced
  • Joint Consensus.
  • Log Compaction.
  • Install Snapshot.

Reference

• Reference1 : https://raft.github.io/raft.pdf
• Reference2 : https://web.stanford.edu/~ouster/cgi-bin/papers/OngaroPhD.pdf
• pseudo code : https://github.com/ongardie/raft-pseudocode

How to Test

• There are 110 test cases. some test cases may be flaky sometimes.
• Test cases in eunit can cover this action
  • RAFT basic.
  • Split Brain
  • Split Vote
  • Joint Consensus
  • Log Compaction
  • Install Snapshot

$ rebar3 eunit
===> Verifying dependencies...
===> Analyzing applications...
===> Compiling raft_erlang_implementation
===> Performing EUnit tests...
..................................................................................................
Finished in 29.611 seconds
110 tests, 0 failures