feat(wasmtime): support wasm components #401

Mossaka · 2023-11-16T04:53:49Z

This PR adds wasm components support to the wasmtime shim which enables it with the component-model feature on. The component loading, linking and execution has a different set of APIs than the wasm module one. I used wasmparser to determine if a byte array is a wasm module or a wasm component, and then dispatch to two different paths.

To finish this work completely, I hope to add some integration tests

add a component test to wasmtime shim

As a follow up PR, I hope to add wasmtime serve functionality to the wasmtime shim so that we have HTTP capability for the shim.

Closes #384

crates/containerd-shim-wasmtime/src/instance.rs

jsturtevant · 2023-11-16T22:09:10Z

Not needed as part of this but we will also need a way to compose together multiple components?

crates/containerd-shim-wasmtime/src/instance.rs

jsturtevant · 2023-11-16T22:08:04Z

crates/containerd-shim-wasmtime/src/instance.rs

+            Some(WasmBinaryType::Component) => {
+                let component = Component::from_binary(&self.engine, &data)?;
+                let file_perms = wasi_preview2::FilePerms::all();
+                let dir_perms = wasi_preview2::DirPerms::all();


I believe this is the way it's done today but in the case of the OCI layers I don't think we need these permissions. Any thoughts on how we can make this more configurable? It feels like this is where a OCI artifact with configuration that had info like this in it would come into play

I guess we need to figure out the story for this.
I can see the argument where te runtime provides an fs component with access to the container root, but you can choose to use your own fs component instead, and ignore/proxy the one provided by the runtime.

If we want to introduce config files as custom layers, it would be good to have a discussion with a wider group (wasi-wg, or even wasi-cg)

If we want to introduce config files as custom layers, it would be good to have a discussion with a wider group (wasi-wg, or even wasi-cg)

They may have already thought of possible solutions, or many people even outside container (e.g., wasmtime itself) could have a similar pain point.

If we want to introduce config files as custom layers, it would be good to have a discussion with a wider group (wasi-wg, or even wasi-cg)

+1 to wider discussion, wasn't intending to figure it out on this PR just calling it out. I don't think it should be a custom layer either, that feels like its too open. I think this is more evidence that we really want a "real" "OCI artifact" with its own custom config media type that contains WASI specific meta data

Raised an issue here for further discussion: #413

crates/containerd-shim-wasmtime/src/instance.rs

Mossaka · 2023-12-04T20:01:18Z

This PR is ready to review. Could you PTAL? @jsturtevant ?

Mossaka · 2023-12-04T20:02:08Z

I am looking for a full fleged wasip2 Command component for testing purposes, but failed to find any existing ones. Asked this question in the Zulipchat: https://bytecodealliance.zulipchat.com/#narrow/stream/219900-wasi/topic/Looking.20for.20a.20wasip2.20component.20for.20testing.20purposes/near/405901766

jsturtevant · 2023-12-04T22:53:48Z

crates/containerd-shim-wasm/src/container/context.rs

+            }
+            Source::Oci([module]) => {
+                log::info!("loading module wasm OCI layers");
+                module.layer.clone()


is the clone required here?

Yes because the function returns a Vec<u8> but module is a &WasmLayer

this was done by design to avoid coping data. Can this return &[u8] instead of vec?

I could roll back to the original one as I am not sure if there is a way to return &[u8] since the file is read in-function.

What do you mean by roll back to the original?

I see the problem, the Source::File would create a temporary variable that can't be returned as borrowed. Maybe need to rethink the source, The read of the file could be moved to the call on entry point and that would make it a non-local variable but this would have cascading changes. Another option might be to have a "intermediate step" that capture that value.

The read of the file could be moved to the call on entrypoint().

I would be a bit more careful here since that means a huge slowdown to entrypoint invocation, which is used in can_handle() and is_linux_container() functions

Reverted the code back.

This definitely looks cleaner than the reverted code, lets revisit this after this PR merges and see if can figure out a way to clean it up

Yup I will have a separate PR to clean that

crates/containerd-shim-wasm/src/container/wasm.rs

crates/containerd-shim-wasmtime/src/instance.rs

jsturtevant · 2023-12-04T23:42:31Z

crates/containerd-shim-wasmtime/src/instance.rs

+    let wasi_data = WasiData {
+        wasi_preview1: wasi_preview1_ctx,
+        wasi_preview2: wasi_preview2_ctx,
+        wasi_preview2_table: wasi_preview2::Table::new(),


should we be going through the effort to build both of these contexts when it is only one type (either module/component)? If not, the build of the store happened insinde the execute_component then this could become something like

fn prepare_wasi_ctx( ctx: &impl RuntimeContext, envs: Vec<(String, String)>, type: wasmbinarytype, ) -> Result<WasiData, anyhow::Error> { match type { Some(WasmBinaryType::Module) => return data { wasi1_data} Some(WasmBinaryType::Component) => return data {wasi2_data},

We could, but I think the current appraoch is nice because we only have one wasi_ctx type that contains both wasip1 and wasip2 APIs. This could potentially benefits a future case where we want to run wasm modules with wasip2 api.

This could potentially benefits a future case where we want to run wasm modules with wasip2 api

what does this look like?

crates/containerd-shim-wasmtime/src/instance.rs

crates/containerd-shim-wasm/src/container/context.rs

Mossaka · 2023-12-05T02:10:10Z

Added a new commit to support wasi:cli/command world that exports a run function.

crates/containerd-shim-wasmtime/src/instance.rs

crates/containerd-shim-wasmtime/src/tests.rs

crates/containerd-shim-wasmtime/src/instance.rs

Signed-off-by: jiaxiao zhou <[email protected]>

This commit adds a simple component testcase to the wasmtime shim in order to test the component model feature. It also refactors some of the common logic to the core crate. It adds a new module named "wasm.rs" to differentiate wasm module from wasm components. Signed-off-by: jiaxiao zhou <[email protected]>

Signed-off-by: jiaxiao zhou <[email protected]>

crates/containerd-shim-wasmtime/README.md

Signed-off-by: jiaxiao zhou <[email protected]>

jsturtevant

LGTM

jsturtevant · 2024-01-11T01:19:21Z

needs some linter fixes

Signed-off-by: jiaxiao zhou <[email protected]>

Mossaka marked this pull request as draft November 16, 2023 04:54

jprendes reviewed Nov 16, 2023

View reviewed changes

crates/containerd-shim-wasmtime/src/instance.rs Outdated Show resolved Hide resolved

Mossaka force-pushed the wasmtime-component-model branch from b2fbb24 to beb52e9 Compare November 16, 2023 18:15

jprendes reviewed Nov 16, 2023

View reviewed changes

crates/containerd-shim-wasmtime/src/instance.rs Outdated Show resolved Hide resolved

jsturtevant reviewed Nov 16, 2023

View reviewed changes

Mossaka mentioned this pull request Dec 4, 2023

Make WASI preopened_dir more configurable #413

Open

Mossaka force-pushed the wasmtime-component-model branch from beb52e9 to eb299ac Compare December 4, 2023 20:00

Mossaka marked this pull request as ready for review December 4, 2023 20:01

Mossaka force-pushed the wasmtime-component-model branch 2 times, most recently from 8418b5b to 795f180 Compare December 4, 2023 23:09

jsturtevant reviewed Dec 4, 2023

View reviewed changes

Mossaka mentioned this pull request Dec 5, 2023

Feature: Support wasi:http/proxy world in wasmtime shim #416

Closed

Mossaka force-pushed the wasmtime-component-model branch from d21141a to 5a133fb Compare December 5, 2023 02:46

jsturtevant reviewed Dec 7, 2023

View reviewed changes

crates/containerd-shim-wasmtime/src/instance.rs Show resolved Hide resolved

jsturtevant reviewed Dec 7, 2023

View reviewed changes

crates/containerd-shim-wasmtime/src/tests.rs Show resolved Hide resolved

jsturtevant reviewed Dec 7, 2023

View reviewed changes

crates/containerd-shim-wasmtime/src/instance.rs Outdated Show resolved Hide resolved

Mossaka force-pushed the wasmtime-component-model branch from 5a133fb to 9f65edb Compare January 3, 2024 21:52

Mossaka requested review from jsturtevant January 3, 2024 22:05

Mossaka added 7 commits January 9, 2024 18:49

feat(wasmtime): support wasm components

1c8b244

Signed-off-by: jiaxiao zhou <[email protected]>

refactor(wasmtime): use upstream wasmparser

f73293b

Signed-off-by: jiaxiao zhou <[email protected]>

feat(wasmtime): add wasi:cli/command support

c810242

Signed-off-by: jiaxiao zhou <[email protected]>

refactor(wasmtime): improve comments

0590fe3

Signed-off-by: jiaxiao zhou <[email protected]>

add documentation for entrypoint

9d755ed

Signed-off-by: jiaxiao zhou <[email protected]>

feat(wasmtime): bump wasmtime to 16

208da25

Signed-off-by: jiaxiao zhou <[email protected]>

Mossaka force-pushed the wasmtime-component-model branch from 76192c4 to 208da25 Compare January 9, 2024 18:50

jsturtevant reviewed Jan 10, 2024

View reviewed changes

crates/containerd-shim-wasmtime/README.md Outdated Show resolved Hide resolved

Mossaka added 3 commits January 10, 2024 23:00

feat(*): revert the change to Source back

0902fea

Signed-off-by: jiaxiao zhou <[email protected]>

docs: makes the screen reader more friendly

bc70968

Signed-off-by: jiaxiao zhou <[email protected]>

docs: comment on how to build the hello world wasm component

b8558fe

Signed-off-by: jiaxiao zhou <[email protected]>

jsturtevant reviewed Jan 11, 2024

View reviewed changes

chore: rustfmt

715b1bb

Signed-off-by: jiaxiao zhou <[email protected]>

Mossaka force-pushed the wasmtime-component-model branch from a1c1987 to 715b1bb Compare January 11, 2024 18:39

Mossaka merged commit 47e2102 into containerd:main Jan 11, 2024
43 checks passed

Mossaka deleted the wasmtime-component-model branch January 11, 2024 20:21

jsturtevant mentioned this pull request Jan 12, 2024

refactor(*): refactor wasm bytes preparation logic to the core crate #448

Closed

andreiltd mentioned this pull request Oct 16, 2024

feat(wasmtime): support wasi:http/proxy world in wasmtime shim #691

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(wasmtime): support wasm components #401

feat(wasmtime): support wasm components #401

Mossaka commented Nov 16, 2023 •

edited

Loading

jsturtevant commented Nov 16, 2023

jsturtevant Nov 16, 2023

jprendes Nov 16, 2023

jprendes Nov 16, 2023

jsturtevant Nov 16, 2023

Mossaka Dec 4, 2023

Mossaka commented Dec 4, 2023

Mossaka commented Dec 4, 2023

jsturtevant Dec 4, 2023

Mossaka Dec 5, 2023

jsturtevant Dec 7, 2023

Mossaka Jan 9, 2024

jsturtevant Jan 10, 2024

Mossaka Jan 10, 2024

Mossaka Jan 10, 2024

jsturtevant Jan 11, 2024

Mossaka Jan 11, 2024

jsturtevant Dec 4, 2023

Mossaka Dec 5, 2023

jsturtevant Dec 7, 2023

Mossaka commented Dec 5, 2023

jsturtevant left a comment

jsturtevant commented Jan 11, 2024

feat(wasmtime): support wasm components #401

feat(wasmtime): support wasm components #401

Conversation

Mossaka commented Nov 16, 2023 • edited Loading

jsturtevant commented Nov 16, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Mossaka commented Dec 4, 2023

Mossaka commented Dec 4, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Mossaka commented Dec 5, 2023

jsturtevant left a comment

Choose a reason for hiding this comment

jsturtevant commented Jan 11, 2024

Mossaka commented Nov 16, 2023 •

edited

Loading