LfrLock: Lock-Free Read Lock

A high-performance Lock-Free Read Lock implementation where reads never block and writes are serialized using a Mutex.

中文文档 | English

Note: If you need a specialized Single-Writer Multiple-Reader (SWMR) version, please use smr-swap directly.

Features

• Lock-Free Reads: Read operations are wait-free and never block, ensuring low latency.
• Serialized Writes: Write operations are serialized using a Mutex to prevent data races.
• Unified Interface: Supports both read and write operations through a single LfrLock<T> type, similar to std::sync::Mutex.
• Easy Usage: Provides a WriteGuard for familiar, mutable access that automatically commits changes on drop.
• Safe Concurrency: Built on top of smr-swap for safe memory reclamation and concurrent access.

No-std Support

lfrlock supports no_std environments. To use it in a no_std crate:

1. Disable default features.
2. Enable the spin feature if you need Mutex support (which LfrLock uses for writes).
3. Ensure alloc is available.

[dependencies]

lfrlock = { version = "0.2", default-features = false, features = ["spin"] }

Note: LfrLock relies on a Mutex for serializing writes. In std environments, it uses std::sync::Mutex. In no_std environments with the spin feature enabled, it uses spin::Mutex.

Quick Start

Installation

Add to your Cargo.toml:

[dependencies]

lfrlock = "0.2"

Basic Usage

use lfrlock::LfrLock;
use std::thread;

#[derive(Debug, Clone)]
struct Data {
    value: i32,
}

fn main() {
    // Create a new LfrLock
    let lock = LfrLock::new(Data { value: 0 });

    let lock_clone = lock.clone();
    let handle = thread::spawn(move || {
        // Read data (never blocks)
        let data = lock_clone.read();
        println!("Reader sees: {}", data.value);
    });

    // Write data using WriteGuard (serialized)
    {
        let mut guard = lock.write();
        guard.value = 42;
    } // Auto-commit on drop

    handle.join().unwrap();
    
    let data = lock.read();
    println!("Final value: {}", data.value);
}

Shared Access (Multi-threaded)

Since LfrLock is not Sync (it contains a thread-local reader), it cannot be shared via Arc<LfrLock>. Instead, obtain a LfrLockFactory from the lock (or create one directly), which is Sync and Clone.

use lfrlock::LfrLock;
use std::sync::Arc;
use std::thread;

fn main() {
    // Create a lock in the main thread
    let lock = LfrLock::new(0);
    
    // Create a factory for sharing (Sync + Clone)
    let factory = lock.factory();
    let factory = Arc::new(factory);

    let mut handles = vec![];

    for i in 0..4 {
        let factory = factory.clone();
        handles.push(thread::spawn(move || {
            // Create a thread-local lock instance
            let lock = factory.create();
            let val = lock.read();
            println!("Thread {} sees: {}", i, *val);
        }));
    }

    // Main thread can still use 'lock'
    lock.store(1);

    for h in handles {
        h.join().unwrap();
    }
}

API Overview

LfrLock<T>

The main type combining reader and writer capabilities.

Creation

• new(initial: T): Creates a new lock with an initial value.
• From<T>: Supports LfrLock::from(value) or value.into().
• Default: When T: Default, supports LfrLock::default().

Read Operations

• read() -> ReadGuard<T>: Gets a lock-free read guard. Never blocks.
• get() -> T: Clones and returns the current value. Requires T: Clone.
• map<F, U>(f: F) -> U: Applies a closure to the current value and returns the transformed result.
• filter<F>(f: F) -> Option<ReadGuard<T>>: Conditional read, returns Some(guard) if closure returns true.
• factory() -> LfrLockFactory<T>: Creates a factory for sharing the lock across threads.

Write Operations

• store(new_value: T): Directly replaces the current value.
• swap(new_value: T) -> T: Atomically swaps and returns the old value. Requires T: Clone.
• update<F>(f: F): Updates data using a closure FnOnce(&T) -> T.
• update_and_fetch<F>(f: F) -> ReadGuard<T>: Updates and returns a guard to the new value.
• fetch_and_update<F>(f: F) -> ReadGuard<T>: Returns a guard to the old value and updates.
• write() -> WriteGuard<T>: Acquires a write lock and returns a guard for mutable access. Requires T: Clone.
• try_write() -> Option<WriteGuard<T>>: Tries to acquire the write lock.

LfrLockFactory<T>

A factory for creating LfrLock instances. Sync and Clone, suitable for sharing across threads.

• new(initial: T): Creates a new factory with an initial value.
• create() -> LfrLock<T>: Creates a new LfrLock handle for the current thread.

WriteGuard<T>

Provides mutable access to the data.

• Automatic Commit: When the guard is dropped, the modified data is atomically swapped in.
• Deref/DerefMut: Access the underlying data transparently.

Implementation Details

LfrLock uses smr-swap internally to manage state. It wraps the Swapper in a Mutex to serialize writes, while the SwapReader allows concurrent, lock-free reads. This design is ideal for read-heavy workloads where writes are infrequent but need to be safe and atomic.

Performance Characteristics

Benchmark results comparing LfrLock against ArcSwap and std::sync::Mutex on an Intel(R) Core(TM) i9-13900KS CPU @ 3.20GHz.

Benchmark Summary

Analysis

• Read Performance: LfrLock provides wait-free reads with nanosecond-scale latency (0.75ns), significantly outperforming ArcSwap and Mutex (~9ns).
• High Contention Reads: In mixed workloads (1:1000 to 1:10 write ratio), LfrLock maintains stable performance (~180-220µs), while ArcSwap degrades significantly at higher write rates (up to ~574µs).
• Write Heavy: Mutex is faster (~18-34%) in pure write-heavy scenarios because LfrLock involves RCU-like operations. ArcSwap is significantly slower.
• Overhead: LfrLock has higher cloning overhead (~94ns) compared to Arc cloning (~9ns) because it registers a new epoch reader. However, it is ~3.2x faster to create than ArcSwap.

License

This project is licensed under either of

• Apache License, Version 2.0, (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.