Using cancelOnGracefulShutdown() causes memory leaks

[xtremekforever]

Consider the following example:

import ServiceLifecycle

var count = 0
let maxCount = 1_000_000

print("Counting up to \(maxCount) elements...")
while count < maxCount {
    // Create an arbitrary stream
    let stream = AsyncStream(unfolding: {
        return Int.random(in: 0..<Int.max)
    })

    // Each call to cancelOnGracefulShutdown() causes a leak
    for await _ in stream.cancelOnGracefulShutdown() {
        count += 1
        break
    }
}

print("Done!")

In here, a new stream is created and destroyed on each loop, then we consume a single element off of the stream (which is just a random number), then exit the stream and do it again.

I am seeing that this code shows that the call to cancelOnGracefulShutdown() causes a memory leak. Here is what I see when I run this example under heaptrack:

Looking at the top-down allocations, it looks like something in the AsyncAlgorithms module is getting allocated and not released:

If I move the creation of the stream out of the while loop and do this:

// Create a stream
let stream = AsyncStream(unfolding: {
    return Int.random(in: 0..<Int.max)
})

while count < maxCount {
    // Each call to cancelOnGracefulShutdown() causes a leak
    for await _ in stream.cancelOnGracefulShutdown() {
        count += 1
        break
    }
}

I still see the exact behavior in heaptrack. It could be that this is a memleak in AsyncAlgorithms itself, but I am not sure based on looking at the output what exactly is causing the leak to occur.

Maybe this is "unintended usage" for cancelOnGracefulShutdown(), but it seems like this could become a problem of an application has to create streams and consume them constantly while still supporting graceful shutdown. Also, this could become a problem for my apps because I use cancelOnGracefulShutdown() everywhere, and having this memory leak is concerning.

[xtremekforever]

Looks like this issue is caused by a memory leak in the Lock struct in swift-async-algorithms. There is a PR here to fix this leak:

apple/swift-async-algorithms#331

I applied the fix locally and re-ran my demo above, and don't get the memory leak anymore: