[BUG] Leaked function closures in alarm invocation

[lukevalenta]

Is there an existing issue for this?

• I have searched the existing issues

What version of workers-rs are you using?

0.6.6

What version of wrangler are you using?

4.38.0

Describe the bug

As mentioned in #722 (comment), there's a slow memory leak that can be triggered by spinning a Durable Object with a periodic alarm. I seem to be able to trigger it both with a 1s alarm (so that the DO does not hibernate) and a 15s alarm (so the DO hibernates in between alarm invocations).

Based on a memory profile taken with devtools (npx wrangler dev and then press d to bring up devtools), it looks like 1-2 kB of function closures are being leaked on each invocation.

In the image below, I ran garbage collection (broom icon) before each memory snapshot. Between snapshots 4 and 5, there was 23.7kB leaked in 154 leaked functions.

I'm been able to observe increasing memory usage both locally with npx wrangler dev and in a deployed worker via internal Cloudflare metrics.

Steps To Reproduce

Code to reproduce is at https://github.com/lukevalenta/workers-rs/tree/f2899fe14a37cd6bce81edc6f08d4f1b7819fec1/do-alarm-memory-leak.

[guybedford]

I can confirm this is a separate leak that still remains, will look into this one further as well - was just working on wasm-bindgen closure internals so it is fresh to mind as well.

[guybedford]

An upstream closure leak bug was found and fixed in wasm-bindgen/wasm-bindgen#4709, once the next wasm-bindgen release goes out it would be worth testing with that to see if that resolves this.