#cache #lru #performance #memory #async

bin+lib tiered-cache

A high-performance multi-tiered cache with automatic sizing

1 unstable release

0.1.6 Nov 8, 2024
0.1.5 Nov 6, 2024

#63 in #lru

Download history 540/week @ 2024-11-04 24/week @ 2024-11-11

564 downloads per month

MIT/Apache

18KB
275 lines

tiered-cache

Crates.io Documentation License

A high-performance multi-tiered cache implementation in Rust with automatic sizing and async support.

Features

  • 🚀 Multiple cache tiers with automatic item placement based on size
  • ⚡ Async support with Tokio
  • 🔄 LRU eviction policy
  • 📊 Detailed statistics for monitoring
  • 🔍 Efficient lookup with DashMap
  • 🛡️ Thread-safe design
  • 📦 Zero unsafe code

Installation

Add this to your Cargo.toml:

[dependencies]
tiered-cache = "0.1.6"

Quick Start

use tiered_cache::{TieredCache, CacheConfig, TierConfig};
use std::time::Duration;

const MB: usize = 1024 * 1024;

// Configure a cache with two tiers
let config = CacheConfig {
    tiers: vec![
        TierConfig {
            total_capacity: 100 * MB,    // 100MB
            size_range: (0, 64 * 1024),  // 0-64KB
        },
        TierConfig {
            total_capacity: 900 * MB,    // 900MB
            size_range: (64 * 1024, MB), // 64KB-1MB
        },
    ],
    update_channel_size: 1024,
};

// Create the cache
let cache = TieredCache::<Vec<u8>, Vec<u8>>::new(config);

// Basic operations
cache.put(b"key1".to_vec(), vec![0; 1024]);
if let Some(value) = cache.get(&b"key1".to_vec()) {
    println!("Retrieved value of size: {}", value.len());
}

// Async get_or_update
let value = cache.get_or_update(b"key2".to_vec(), async {
    None // Returns None instead of Some(vec![0; 2048])
}).await;

Cache Configuration

The cache is configured using tiers, where each tier has:

  • A total capacity in bytes
  • A size range for entries (min_size, max_size)

Items are automatically placed in the appropriate tier based on their size. The cache uses the HeapSize trait to accurately measure memory usage.

Performance

The implementation uses:

  • DashMap for concurrent key-to-tier mapping
  • parking_lot locks for low contention
  • SmallVec for efficient tier storage
  • Cache-padded structures to prevent false sharing

Examples

See the examples directory for more usage examples, including:

  • Setting up a 1GB cache with multiple tiers
  • Handling different value sizes
  • Monitoring cache statistics

Note

The update_channel_size: 1024 is used to configure the capacity of a broadcast channel that notifies subscribers about cache updates. This is implemented using Tokio's broadcast channel.

In the code, specifically in lib.rs, we can see:

let (tx, _) = broadcast::channel(config.update_channel_size);

This channel is used to notify interested parties when values in the cache are updated. Users of the cache can subscribe to these updates using the subscribe_updates() method:

/// Subscribes to cache updates
#[inline]
pub fn subscribe_updates(&self) -> broadcast::Receiver<K> {
    self.update_tx.subscribe()
}

When values are updated in the cache (specifically in the update_value method), notifications are sent through this channel:

#[inline]
fn notify_update(&self, key: K) {
    let _ = self.update_tx.send(key);
}

The size of 1024 means that the channel can buffer up to 1024 update notifications before older messages start getting dropped. This is useful in scenarios where you want to monitor or react to cache updates, such as:

  • Synchronizing multiple cache instances
  • Logging cache operations
  • Triggering side effects when cache entries are updated

If you expect a very high rate of cache updates, you might want to increase this value. Conversely, if you don't need update notifications, you could set it to a smaller value to save memory.

License

Licensed under either of:

at your option.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Dependencies

~5–12MB
~126K SLoC