RaftMessaging - A Raft-Based Application for Group Communication (Not Encrypted)

Abstract

A Java 21+ Proof-of-Concept for a message streaming distributed that ensurens data consistency through the Raft consensus algorithm.

Documentation

This project implements the Raft Consensus Algorithm, designed to manage a consistent and fault-tolerant distributed state. The system facilitates a multi-room chat application where messages are atomically replicated across a cluster of independent nodes.

A Technical Report is available in this repo.

There is also a Written Integration which aims to illustrate a bit deeper some choices about the project.

System Architecture

The core logic resides in the com.raft.node package. Each node operates as a finite state machine that can transition between three roles: Follower, Candidate, and Leader.

Key Components

• Node.java: Manages business logic, state transitions, and integration with networking and storage subsystems.
• Replicated Log: A strictly ordered sequence of LogEntry objects containing commands to be applied to the state machine.
• WalStorage.java: Implements persistence using Write-Ahead Logging to ensure data durability across crashes.
• Network Abstraction: Support for both InMemoryNetwork.java (for localized testing) and HttpNetwork.java (for real-world HTTP-based deployments).

Protocol Features

The implementation adheres to the original Raft specification while incorporating optimizations for high-performance scenarios:

1. Leader Election: Utilizes randomized timeouts to minimize election collisions and ensure cluster stability.
2. Pre-Vote Phase: A safety optimization that prevents unnecessary term inflation by checking candidacy eligibility before incrementing the local term.
3. Log Compaction (Snapshotting): An asynchronous mechanism that prunes the replicated log and saves the current state machine, preventing infinite memory growth.
4. Idempotency: Client session tracking ensures exactly-once semantics for commands, even during network retries or leader changes.
5. Natural Batching: Optimizes network throughput by aggregating multiple log entries into a single AppendEntries RPC.

Getting Started

Prerequisites

• Java 21+ (utilizing Virtual Threads/Project Loom).
• Maven 3.8+.
• Docker & Docker Compose (optional for containerization).

Web Interface

An interactive monitoring dashboard (index.html) is provided to:

• Monitor real-time log synchronization across the cluster.
• Send messages to specific rooms via node-specific HTTP gateways.
• Observe system resilience by dynamically switching the gateway node.

In order to access to the web interface, you need to start the docker service with:

docker-compose up --build

To clear the old volumes and images, please use:

docker-compose down -v

docker builder prune -f

Performance Analysis

By leveraging Virtual Threads, the system manages high concurrency with minimal CPU overhead. The batching mechanism allows for performance peaks exceeding 15,000 requests per second during in-memory stress testing.