raft-kotlin

Kotlin implementation of the raft consensus algorithm

A raft is a consensus algorithm that is designed to be easy to understand. It's equivalent to Paxos in fault-tolerance and performance

Repository provides an example implementation and show-case usage with in-memory key-value storage

👨‍💻 How to run (key-value example)

#Cloning repository
git clone https://github.com/AChepurnoi/raft-kotlin.git

#Building jar file
./gradlew jar

docker-compose up --build

To interact with cluster you can use cURL:

#Set test=hello
docker run --rm --net=raft-kt_default hortonworks/alpine-curl:3.1 curl --request POST --url node_one:8000/test --data 'hello' ; echo

#Read test value
docker run --rm --net=raft-kt_default hortonworks/alpine-curl:3.1 curl --request GET --url node_one:8000/test ; echo

The key-value implementation uses 307 Redirect to redirect requests from slaves to the master.

This requires you to be able to resolve the IP from configuration (You should interact with HTTP server only from docker network e.g. you container)

Another option is to run jar files locally with proper env configuration

To read node list from env, raft env configuration uses the following notation:

NODES=[ID]:[HOST]:[PORT],[ID]:[HOST]:[PORT]...

Shows the example of how raft module can be used to implement distributed in-memory key-value storage.

Current implementation exposes two endpoints:

# Set `key={request_body}
POST HOST/{key} 

#Returns the value of the `key` or `Nil` if the key does not exist
GET HOST/{key}

Key-value HTTP server uses 8000 port by default

Exposes RaftNode class for clients to create a cluster node, actions to mutate the state of the cluster and method to fetch the current state.

Components:

Unit tests - actions, clock, log, state and RaftController classes are tested
Integration tests - RaftClusterTesting class contains different test cases for living cluster (With LocalRaftNode instead of GrpcClusterNode)
Key-Value container testing - kv cluster testing - Not implemented yet

This is not production ready implementation and very likely there are bugs