pbft-rs

Practical Byzantine Fault Tolerance in Rust

Algorithm Overview

Castro and Liskov's pBFT provides Byzantine fault tolerance for state machine replication, tolerating up to f < n/3 faulty nodes through a three-phase protocol.

Key Implementation Components

Three-Phase Consensus Protocol:

pub enum PbftMessage {
    PrePrepare(PrePrepareMessage),
    Prepare(PrepareMessage), 
    Commit(CommitMessage),
    ViewChange(ViewChangeMessage),
    NewView(NewViewMessage),
}

Pre-prepare Phase: Primary orders requests and broadcasts pre-prepare messages Prepare Phase: Replicas validate and broadcast prepare messages
Commit Phase: After 2f+1 matching prepares, replicas broadcast commits

View Change Protocol:

• Timeout-based primary failure detection
• View change messages with prepared request proofs
• New primary computes new view message with complete pre-prepare set

Async/Concurrent Architecture:

struct PbftConsensus {
    state: Arc<RwLock<PbftNode>>,
    network: NetworkManager,
    service: Box<dyn Service>,
}

Production Features:

• Tokio-based async networking
• Message authentication with HMAC-SHA256
• Efficient batching and pipelining
• Read-only request optimization
• Checkpoint and garbage collection
• Persistent state management

Performance Optimizations:

• Request batching reduces consensus overhead
• Parallel signature verification
• Memory pool management for frequent allocations
• Zero-copy message processing where possible