Distributed Consensus Gym

A reinforcement learning environment for studying distributed consensus algorithms with fault tolerance.

Overview

This gym environment simulates a distributed system where agents must reach consensus while handling various failure modes including Byzantine faults, network partitions, and node crashes.

Features

• Multi-agent RL Environment: Compatible with OpenAI Gym
• Network Simulation: Realistic network delays, packet loss, and partitions
• Fault Injection: Byzantine nodes, crashes, and slow nodes
• Consensus Algorithms: Pluggable consensus implementations
• Training Utilities: Multi-agent training with fault scenarios

Quick Start

pip install -r requirements.txt
python examples/basic_training.py

Usage Examples

Basic Training

import gym
import consensus_gym

env = gym.make('ConsensusEnv-v0', num_nodes=5)
obs = env.reset()

for step in range(1000):
    actions = {i: env.action_space.sample() for i in range(env.num_nodes)}
    obs, rewards, done, info = env.step(actions)
    if done:
        print(f"Consensus reached: {info['consensus_value']}")
        break

Network Fault Testing

from consensus_gym.utils.training import FaultInjector

fault_injector = FaultInjector()
fault_injector.inject_fault(env, 'byzantine', [0, 1])
env.network.create_partition([2, 3])

Environment Details

Observation Space

• Node state vectors
• Message queues
• Network topology

Action Space

• Consensus proposals
• Message broadcasts
• Vote decisions

Reward Structure

• Consensus achievement: +10
• Time penalty: -0.1 per step
• Byzantine behavior: -5

Configuration

See config.py for environment parameters:

• Network latency and loss rates
• Byzantine fault ratios
• Training hyperparameters