#ring-buffer #lock-free #disruptor #high-performance #event-processing

disruptor-rs

A high-performance ring buffer implementation of the LMAX Disruptor pattern

2 releases

0.1.1 Nov 15, 2024
0.1.0 Nov 2, 2024

#253 in Concurrency

Download history 119/week @ 2024-10-28 34/week @ 2024-11-04 105/week @ 2024-11-11 24/week @ 2024-11-18 4/week @ 2024-11-25 4/week @ 2024-12-02 18/week @ 2024-12-09

57 downloads per month
Used in onering

MIT/Apache

81KB
1.5K SLoC

Disruptor-rs

Build Status Code Coverage Crates.io Documentation License Rust Version

A heavily documented Rust implementation of the LMAX Disruptor pattern focused on high-performance inter-thread messaging.

Origins & Purpose

This project is a fork of disrustor by Sebastian Klose. While the original implementation provided an excellent foundation, this fork aims to:

  1. Provide comprehensive documentation explaining the Disruptor pattern
  2. Rewrite the implementation in a more idiomatic Rust style
  3. Create a production-ready, Rust-native package for the community

Important Differences from the original implementation

  • The most important difference is that this implementation is that the Sequencer uses WaitingStrategy to get coordinate between the sequences instead of the default implied busy-spinning. This makes the disruptor more consistent and potentially more performant and efficient in its CPU usage.

What is the Disruptor?

The Disruptor is a high-performance inter-thread messaging library, originally developed by LMAX Exchange for their financial trading platform. It achieves superior performance through:

  • Lock-free algorithms
  • Cache-line padding to prevent false sharing
  • Efficient memory pre-allocation
  • Mechanical sympathy with modern CPU architectures

Key Components

1. Ring Buffer

A fixed-size circular buffer that pre-allocates memory for events:

  • Zero garbage collection
  • Cache-friendly memory access patterns
  • Power-of-2 sizing for efficient modulo operations

2. Sequences

Cache-line padded atomic counters that track:

  • Producer position in the ring buffer
  • Consumer progress
  • Dependencies between consumers

3. Sequencer

Coordinates access to the ring buffer:

  • Manages sequence claim and publication
  • Implements backpressure
  • Ensures thread-safety

4. Event Processing

  • Producers: Write events to the ring buffer
  • Consumers: Process events through user-defined handlers
  • Barriers: Coordinate dependencies between processors

5. Waiting Strategies

Different strategies for thread coordination:

  • BusySpinWaitStrategy: Lowest latency, highest CPU usage
  • YieldingWaitStrategy: Balanced approach
  • SleepingWaitStrategy: Lowest CPU usage, higher latency

Usage Example

use disruptor_rs::{DisruptorBuilder, EventHandler};
// Define your event type
#[derive(Default)]
struct MyEvent {
    data: i64,
}
// Define your event handler
#[derive(Default)]
struct MyHandler;
impl EventHandler<MyEvent> for MyHandler {
    fn on_event(&self, event: &MyEvent, sequence: i64, end_of_batch: bool) {
        println!("Processing event: {} at sequence {}", event.data, sequence);
    }
}
// Build and run the disruptor
let (executor, producer) = DisruptorBuilder::with_ring_buffer::<MyEvent>(1024)
    .with_busy_spin_waiting_strategy()
    .with_single_producer_sequencer()
    .with_barrier(|scope| {
        scope.handle_events(MyHandler::default());
    })
    .build();
// Start processing
let handle = executor.spawn();
// Produce events
// ...
// Cleanup
producer.drain();
handle.join();

Performance Considerations

  • Ring Buffer Size: Must be a power of 2
  • Waiting Strategy: Choose based on your latency/CPU trade-offs
  • Event Handlers: Keep processing logic lightweight
  • Batch Processing: Use batch writes when possible

Contributing

Contributions are welcome! Please feel free to submit a Pull Request. Areas of interest:

  • Additional waiting strategies
  • Performance optimizations
  • Documentation improvements
  • Example use cases

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

Dependencies

~8–18MB
~250K SLoC