created_at2021-11-22 07:01:18.240907
updated_at2021-11-22 07:01:18.240907
descriptionMinimal async cache in Rust with support for key expirations
gcxfd (gcxfd)




Build Status Crates.io

This crate offers a very small cache with asynchronous bindings, allowing it to be used in async Rust contexts (Tokio, async-std, smol, etc.) without blocking the worker thread completely.

It also includes the ability to expire entries in the cache based on their time inside; this is done by spawning a monitor on your async runtime in order to perform cleanup tasks periodically. The eviction algorithm is similar to the one found inside Redis, although keys are not removed on access in order to reduce borrow complexity.

This crate is still a work in progress, so feel free to file any suggestions or improvements and I'll get to them as soon as possible :).

Getting Started

This crate is available on crates.io. The easiest way to use it is to add an entry to your Cargo.toml defining the dependency:

retainer = "0.2"

Basic Usage

The construction of a cache is very simple, and (currently) requires no options. If you need to make use of key expiration, you must ensure to either await a monitor or spawn a monitor on your runtime.

There are many ways to provide an expiration time when inserting into a cache, by making use of several types implementing the Into<CacheExpiration> trait. Below are some examples of types which are available and some of the typical APIs you will find yourself using. This code uses the Tokio runtime, but this crate should be compatible with most of the popular asynchronous runtimes. Currently a small set of tests are run against async-std, smol and Tokio.

use retainer::Cache;
use tokio::time::sleep;

use std::sync::Arc;
use std::time::{Duration, Instant};

async fn main() {
    // construct our cache
    let cache = Arc::new(Cache::new());
    let clone = cache.clone();

    // don't forget to monitor your cache to evict entries
    let monitor = tokio::spawn(async move {
        clone.monitor(4, 0.25, Duration::from_secs(3)).await

    // insert using an `Instant` type to specify expiration
    cache.insert("one", 1usize, Instant::now()).await;

    // insert using a `Duration` type to wait before expiration
    cache.insert("two", 2, Duration::from_secs(2)).await;

    // insert using a number of milliseconds
    cache.insert("three", 3, 3500).await;

    // insert using a random number of milliseconds
    cache.insert("four", 4, 3500..5000).await;

    // insert without expiration (i.e. manual removal)
    cache.insert_untracked("five", 5).await;

    // wait until the monitor has run once

    // the first two keys should have been removed

    // the rest should be there still for now

    // wait until the monitor has run again

    // the other two keys should have been removed

    // the key with no expiration should still exist

    // but we should be able to manually remove it

    // and now our cache should be empty

    // shutdown monitor

In the case this example is not kept up to date, you can look for any types which implement the Into<CacheExpiratio> trait in the documentation for a complete list.

Cache Monitoring

All key expiration is done on an interval, carried out when you await the future returned by Cache::monitor. The basis for how this is done has been lifted roughly from the implementation found inside Redis, as it's simple but still works well.

When you call Cache::monitor, you need to provide 3 arguments:

  • sample
  • frequency
  • threshold

Below is a summarization of the flow of eviction, hopefully in a clear way:

  1. Wait until the next tick of frequency.
  2. Take a batch of sample entries from the cache at random.
  3. Check for and remove any expired entries found in the batch.
  4. If more than threshold percent of the entries in the batch were removed, immediately goto #2, else goto #1.

This allows the user to control the aggressiveness of eviction quite effectively, by tweaking the threshold and frequency values. Naturally a cache uses more memory on average the higher your threshold is, so please do keep this in mind.

Cache Logging

As of v0.2, minimal logging is included using the log crate. You can attach any of the compatible logging backends to see what is happening in the cache (particularly the eviction loop) to better gauge your usage and parameters.

Commit count: 64

cargo fmt