oraft consists of:
-
a core Raft (purely functional) state machine featuring leader election + log replication, log compaction via snapshotting, cluster membership change support and efficient linearizable read-only operations. This state machine is abstracted over the specifics of peer communication (I/O, message serialization), concurrency and timeouts. (
Oraftmodule) -
a specialization of the above state machine using the Lwt library for concurrency (
Oraft_lwtmodule) -
a replicated state machine built atop
Oraft_lwt(RSMmodule) -
a sample distributed dictionary client/server (
dict.ml) built atopRSM
The core state machine has been tested using a discrete event simulator that simulates node failures, network failures, message loss, random delays, cluster changes (node deployment and decommissioning)...
The sample distributed dictionary has been clocked at rates exceeding 70000 ops/s on a 3-node cluster. There is potential for optimization, both in the core state machine (by decreasing GC pressure via lesser copying) and the communication layer (faster message serialization, command batching).
In Search of an Understandable Consensus Algorithm. Diego Ongaro and John Ousterhout. Stanford University. (Draft of October 7, 2013). https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf