#ring-buffer #locking #zig #read #sequentially #shared #thread

zsling

Rust Wrapper around a Sequentially lockign (SeqLock) Ring Buffer written in Zig

4 releases

0.1.2 Feb 19, 2023
0.1.1 Feb 2, 2023
0.1.0 Jan 31, 2023
0.0.1 Jan 31, 2023

#1205 in Concurrency

MIT license

34KB
632 lines

This crates provides a sequentially locking Ring Buffer. It allows for a fast and non-writer-blocking SPMC-queue, where all consumers read all messages.

Original

This is the Rust-wrapped Zig version of the sling crate. Note that due to current constraints in the implementation, the buffer size is set to 256 and the and messages are set to [u8; 8].

Usage

There are two ways of consuming from the queue. If threads share a SharedReader through a shared reference, they will steal queue items from one anothers such that no two threads will read the same message. When a SharedReader is cloned, the new SharedReader's reading progress will no longer affect the other one. If two threads each use a separate SharedReader, they will be able to read the same messages.

# use zsling::*;

let buffer = RingBuffer::new();

let mut writer = buffer.try_lock().unwrap();
let mut reader = buffer.reader();

std::thread::scope(|s| {
    let reader = &reader;
    for t in 0..8 {
        s.spawn(move || {
            for _ in 0..100 {
                if let Some(val) = reader.pop_front() {
                    println!("t: {}, val: {:?}", t, val);
                };
            }
        });
    }

    for i in 0..100 {
        writer.push_back([0, 1, 2, 3, 4, 5, 6, 7]);
    }
});

Important!

It is also important to keep in mind, that slow readers will be overrun by the writer if they do not consume messages quickly enough. This can happen quite frequently if the buffer size is not large enough. It is advisable to test applications on a case-by-case basis and find a buffer size that is optimal to your use-case.

Dependencies

~93KB