It seems they proposed a way to resolve the contradiction I raised (io_uring isn't supported on cloud providers) with the ground reality in the company's blog post.
Namely, that:
* it is intended for edge function deployment.
* the paper they mention that is informing their decision to rewrite is based on perf improvements in the longtail by using io_uring.
I framed it as "Are we sure...?", but truth is, I know providers don't have io_uring support. This is relatively straightforward to derive from edge functions are well-known to be multi-tenant.
A cursory search shows unresolved AWS tickets re: this, and multiple announcements from Google about how it has been disabled in gCloud.
Thus, it is likely they understood I was framing it politely, and that there was a clear contradiction here, hence their reply to me, raising a potential resolution for that contradiction, a resolution I hadn't considered.
I don't see anything complaining, or bringing up, or implicitly denying, all the other stuff ("open source", "MIT licensed", "open contribution", "software for free" "under no obligation to use."). In fact, they explicitly indicated they completely agree with that view ("it's very cool that they're making parts of their tech FOSS.")
The Turso company provisions infrastructure using fly.io, which uses firecracker VM (which as of now does support io_uring afaict). They are using the term “serverless” liberally maybe? Goodness knows we all do.
io_uring isn't supported on cloud providers for now, but it may not be the case in future. Which is fine, because Limbo won't be ready for a few years at least.
It seems they proposed a way to resolve the contradiction I raised (io_uring isn't supported on cloud providers) with the ground reality in the company's blog post.
Namely, that:
* it is intended for edge function deployment.
* the paper they mention that is informing their decision to rewrite is based on perf improvements in the longtail by using io_uring.
I framed it as "Are we sure...?", but truth is, I know providers don't have io_uring support. This is relatively straightforward to derive from edge functions are well-known to be multi-tenant.
A cursory search shows unresolved AWS tickets re: this, and multiple announcements from Google about how it has been disabled in gCloud.
Thus, it is likely they understood I was framing it politely, and that there was a clear contradiction here, hence their reply to me, raising a potential resolution for that contradiction, a resolution I hadn't considered.
I don't see anything complaining, or bringing up, or implicitly denying, all the other stuff ("open source", "MIT licensed", "open contribution", "software for free" "under no obligation to use."). In fact, they explicitly indicated they completely agree with that view ("it's very cool that they're making parts of their tech FOSS.")