Stream: wasmtime

Topic: embedders: who is using call_hook and where?


view this post on Zulip Pat Hickey (Jan 28 2026 at 21:59):

I introduced the call_hook machinery when I was working on Fastly's embedding of wasmtime. I'm noodling what some design changes there might look like and I'm curious if any other embedders are actually using that feature with components, and if so, what are you using it for?

view this post on Zulip Pat Hickey (Jan 28 2026 at 22:00):

@Dan Gohman is Fastly still using the call hook in roughly the same way as before with the component embedding?

view this post on Zulip Pat Hickey (Jan 28 2026 at 22:01):

pinging some other folks who have maintained major embeddings: @Joel Dice @Bailey Hayes @Yosh Wuyts @Saúl Cabrera

view this post on Zulip Dan Gohman (Jan 28 2026 at 22:07):

@Pat Hickey We're still using it, or rather, we're using call_hook_async now, but for effectively the same purpose, iiuc.

view this post on Zulip Bailey Hayes (Jan 28 2026 at 22:13):

we're not currently using it

view this post on Zulip Yosh Wuyts (Jan 28 2026 at 23:41):

cc/ @Dan Chiarlone (danbugs) @Lucy Menon

view this post on Zulip Saúl Cabrera (Jan 28 2026 at 23:57):

cc/ @Jeff Charles

view this post on Zulip Alex Crichton (Jan 29 2026 at 03:45):

I believe that Spin is using the synchronous version for tracking time-in and time-out for requests

view this post on Zulip Till Schneidereit (Jan 29 2026 at 14:30):

correct yeah, we're using it. However, I think we should actually do more coarse-grained time tracking based on Tokio tasks, so if there's a good argument for removing them, that's something we could discuss

view this post on Zulip Jeff Charles (Jan 29 2026 at 15:36):

We're not currently using it

view this post on Zulip Lann Martin (Jan 30 2026 at 16:01):

Spin uses it to emit metrics for guest execution wall time. Till is right that we could potentially replace that with task busy time but they aren't exactly equivalent. We've also discussed using them for tracking CPU counters but I'd be even happier to add e.g. USDT probes for that (since counters aren't terribly portable to begin with).

view this post on Zulip Lucy Menon (Feb 05 2026 at 14:30):

I don't believe we are using it right now. We don't need per-instance execution time or anything like that at the moment.

view this post on Zulip Pat Hickey (Feb 05 2026 at 16:43):

Thanks all. FWIW I moved on to working on another mechanism at the moment and don't plan to make any changes to call hooks.


Last updated: Feb 24 2026 at 04:36 UTC