Crate concurrent_lru

An implementation of a concurrent LRU cache. It is designed to hold heavyweight resources, e.g. file descriptors, disk pages. The implementation is heavily influenced by the [LRU cache in LevelDB].

Currently there are two implementations, unsharded and sharded.

• unsharded is a linked hashmap protected by a big lock.
• sharded shards unsharded by key, providing better performance under contention.

Example

use concurrent_lru::sharded::LruCache;
use std::{fs, io};

fn read(_f: &fs::File) -> io::Result<()> {
    // Maybe some positioned read...
    Ok(())
}

fn main() -> io::Result<()> {
    let cache = LruCache::<String, fs::File>::new(10);

    let foo_handle = cache.get_or_try_init("foo".to_string(), 1, |name| {
        fs::OpenOptions::new().read(true).open(name)
    })?;
    read(foo_handle.value())?;
    drop(foo_handle); // Unpin foo file.

    // Foo is in the cache.
    assert!(cache.get("foo".to_string()).is_some());

    // Evict foo manually.
    cache.prune();
    assert!(cache.get("foo".to_string()).is_none());

    Ok(())
}

Modules

sharded

unsharded