⏱️ At a hostcall to get the amount of vcpu time that has passed in the guest, in milliseconds #412
Conversation
crates/adapter/src/fastly/core.rs
Outdated
| @@ -268,6 +268,24 @@ pub mod fastly_abi { | |||
| } | |||
| } | |||
|
|
|||
| pub mod fastly_vcpu { | |||
There was a problem hiding this comment.
drive-by feedback: might be worth thinking whether to use a more generic module name where we can add other bits of introspection. E.g.: fastly_introspect or fastly_meta or similar...
There was a problem hiding this comment.
(I can imagine a lot of additional introspection we might want to add over time, on other aspects of the guest behavior.)
There was a problem hiding this comment.
I went with fastly_compute_runtime, which I like as a general future module for getting or setting or otherwise looking at the behavior of a session itself.
JakeChampion
force-pushed
the
awick/get-vcpu-ms
branch
from
7ff327a
to
596e5eb
Compare
Co-authored-by: Cameron Walters (cee-dub) <[email protected]>
There was a problem hiding this comment.
Did you rerun make adapter after merging main so that this includes both the vcpu hostcall and the inspect hostcall?
There was a problem hiding this comment.
Yup! Rerunning it now shows me no changes with this PR.
This is a proposed extension, to help customers diagnose where their code is spending time.
One explicit thing to mention is that while this hostcall limits itself to vcpu time (and thus does not count any time spent blocking on IO, for example), it is based on a time passing rather than cycles, WASM instruction count, or any other timer that is independent of the underlying processor(s) running the code. Thus, the values found should only really be compared to other values captured in the exact same run.