A tiny framework for managing resources.
I built this while working with mediasoup, a library for creating video conferencing apps. It uses a lot of very nested resource handles, and if you forget to release one of them, it'll eventually crash your server and kill active calls.
This library manages resource hierarchies in a RAII-inspired harness. Prune one branch and everything it allocated is released in the correct order. No resource leaks, no crashed servers, no unhappy callers.
It works by wrapping objects in Proxy.revocable(...) with destructor hooks. While great for guaranteeing resources are deconstructed and garbage collected, you might get opaque use-after-free errors. Basically: null pointer exceptions.
Easier to handle than resource exhaustion, but the abstraction isn't free.
Here's an example: let's say you've got a thread pool, one per CPU. Each thread gets a Resource, a small wrapper that hooks into setup and teardown controls.
async function Worker() {
const thread = await spawn();
return {
// The value returned after initialization completes
value: thread,
// Called when the resource is destroyed
destroy: () => thread.close(),
};
}Now define a pool that creates and manages workers:
async function WorkerPool(
{ create }: ResourceScope,
config: { poolSize: number },
) {
const promises = Array(config.poolSize).fill(Worker).map(create);
const threads = await Promise.all(promises);
return {
// ... External API goes here ...
value: {
doSomeWork() {},
doSomethingElse() {},
},
};
}Finally, create the pool:
const pool = await create(WorkerPool, {
poolSize: cpus().length,
});
// Provisioned and ready to go!
pool.doSomeWork();
pool.doSomethingElse();The magic of this framework is that resources never outlive their owners. If you tear down the pool, it will deallocate everything beneath it first:
await destroy(pool);
// [info] closing worker
// [info] closing worker
// [info] closing worker
// [info] closing worker
// [info] closing poolNo more forgotten resources.
The framework can be used to manage small pieces of stateful logic in your application, or it can scale to manage your entire server. Use the paradigm as much or as little as you like.
I built this for my own projects. Documentation is a bit sparse, but enough GitHub stars could change that. This is a bribe.