A

Raft Implementation

Tugas Besar 1 Sistem Paralel dan Terdistribusi Kelompok [A]

Oleh Kelompok A:

1. 13522020 Aurelius Justin Philo Fanjaya
   
2. 13522090 Fedrianz Dharma
   
3. 13522094 Andhika Tantyo Anugrah
   

Sekolah Teknik Elektro dan Informatika
Institut Teknologi Bandung
Semester VI Tahun 2024/2025

Overview

This project implements the Raft consensus algorithm in Java using gRPC for communication. The implementation provides a distributed key-value store with the following features:

Raft Features Implemented

• Heartbeat: Node health monitoring & periodic messages
• Leader Election: Leader node failover mechanism
• Log Replication: Cluster action logging system
• Membership Change: Mechanism to add and remove nodes from the cluster
• gRPC Communication: Remote Procedure Call protocol for inter-node communication

Key-Value Store Operations

• ping - Check connection with server (returns "PONG")
• get <key> - Get value for a key (returns empty string if key doesn't exist)
• set <key> <value> - Set value for a key (overwrites if exists)
• strln <key> - Get length of value for a key
• del <key> - Delete entry for a key (returns deleted value)
• append <key> <value> - Append to value of a key (creates key if doesn't exist)

Dependencies

The project uses Maven for dependency management. Required dependencies are automatically managed through pom.xml:

• gRPC Java (1.72.0)
• Protocol Buffers (3.25.1)
• Gson (2.13.1)

Building the Project

mvn clean compile

Running the System

Starting Servers

Start the first server (becomes leader):

java -cp target/classes Main server localhost 8080

Start additional servers (join existing cluster):

java -cp target/classes Main server localhost 8081 localhost 8080
java -cp target/classes Main server localhost 8082 localhost 8080

Running Client

Start the client with known server addresses:

java -cp target/classes Main client localhost:8080 localhost:8081 localhost:8082

Client Commands

Once the client is running, you can use these commands:

> ping
PONG

> set key1 value1
OK

> get key1
"value1"

> append key1 value2
OK

> get key1
"value1value2"

> strln key1
12

> del key1
"value1value2"

> get key1
""

> log
Leader Log:
  [0] Term:1 Command:set key1 value1
  [1] Term:1 Command:append key1 value2
  [2] Term:1 Command:del key1

> quit

Architecture

Server Components

• RaftNode: Implements the Raft consensus algorithm
• KVStore: In-memory key-value storage application
• RaftServer: gRPC server wrapper

Client Components

• RaftClient: gRPC client with leader discovery and redirection

gRPC Services

• execute: Execute application commands (client interface)
• requestLog: Get leader's log entries (client interface)
• applyMembership: Add new nodes to cluster (client interface)
• appendEntries: Log replication (inter-node)
• requestVote: Leader election (inter-node)
• heartbeat: Health monitoring (inter-node)

Consensus Properties

The implementation ensures:

• Consistency: All nodes agree on the same log
• Partition Tolerance: System continues to operate despite network partitions
• Majority Requirement: Commands are only committed when replicated to majority of nodes (⌊n/2⌋ + 1)
• Leader Redirection: Non-leader nodes redirect clients to the current leader

Fault Tolerance

• Leader Failure: Automatic leader election when heartbeats are missed
• Node Failure: Failed nodes are detected but not automatically removed from cluster
• Network Partitions: Minority partitions cannot commit new entries
• Split Brain Prevention: Only majority partitions can elect leaders

Testing

You can test the system by:

1. Starting multiple servers
2. Using the client to perform operations
3. Stopping the leader server to test election
4. Checking log consistency across nodes

Links

• Tugas Besar 1 IF3130 - Sistem Paralel dan Terdistribusi
• This Repository