Conversation
dicej
left a comment
There was a problem hiding this comment.
Looks great; thanks for doing this! Just a couple of optional comments inline.
This reminds me that we need still need to nail down the "preventing instance reuse" story in Spin and Wasmtime. For example, we've discussed using backpressure_inc in combination with graceful shutdown ending with a call to wasi:cli/exit#exit, but I'm not sure Spin is quite ready for that yet; it might log an alarming message when the instance exits, failing to distinguish it from other kinds of traps. And in any case, we should document this somewhere upstream (e.g. in the wasi-http spec) before we declare the Spin implementation finished.
I'll open an issue on the WASI repo.
content/v3/http-trigger.md
Outdated
|
|
||
| The exact details depend on your component. | ||
|
|
||
| A WASI P2 component is _never_ subject to instance reuse. (This includes all triggers other than HTTP.) |
There was a problem hiding this comment.
We could mention here that future versions of Spin might support optionally enabling consecutive reuse for P2 components.
There was a problem hiding this comment.
I noodled around with some words for this but none of them felt very actionable to me, so I'm going to pass on this for now. I've just softened "never" to "not" to avoid any implication that P2 components will never ever ever see reuse.
|
Per my comment above, I've opened an issue about documenting instance reuse on the WASI repo: WebAssembly/WASI#898 |
Signed-off-by: itowlson <ivan.towlson@fermyon.com>
|
Talked to @mikkelhegn and we might want to hold off on these until we fork the 4.x docs and can have them be all-P3, instead of awkwardly mixing P2 and P3 docs. |
|
Folded this into #183 |
I'm going to drop these bits out in multiple commits with a view to making them easier to review, although I'm pretty sure my laziness with git will thwart that noble aim before too long.
So this initial commit is just the instance reuse stuff and I'll deal to other bits over time.