To create the wasm:
py2wasm hello-wasm-world.py -o hello-wasm-world.wasm
To run the wasm:
wasmer run hello-wasm-world.wasm
Instructions for installing wasmer: https://docs.wasmer.io/install
Why?
Docker and WASM are different layers of abstraction. If Docker is a "container for the OS", then WASM would be a "container for the application". That leads to the conclusion that you can prescind from Docker altogether, theoretically. However, in the real world, there are more factors at play: orchestration and scaling (there's still no k8s equivalent for wasm, afaik), developer experience, maturity (the wasm ecosystem is relatively young), cost of change, and "politics". So, I think as of now, if some organization, that is already running micr
6690
oservices, were to use WASM, it would do it with Docker.
How does Docker work with WASM?
cli <-> dockerd <-> containerd <-> shim <-> wasm rt (replacing runc) <-> registry
To create the wasm docker image:
docker build -t kcandidate/hello-wasm-world:v1 --platform wasi/wasm32 .
To use a wasm runtime for docker, follow these instructions:
https://docs.docker.com/engine/daemon/alternative-runtimes/.
But those instructions are incomplete (as of 2024/12/13).
The daemon configuration file is /etc/docker/daemon.json:
{
"features": {
"containerd-snapshotter": true
}
}
To build the wasmtime binary, you'll need protobuf-compiler and libseccomp-dev which are not present in the build stage image used in the Docker docs. Alternatively, you can just get the precompiled binary from https://github.com/containerd/runwasi/releases
docker build --output . - <<EOF
FROM rust:latest as build
RUN apt update && apt install -y protobuf-compiler libseccomp-dev
RUN cargo install \
--git https://github.com/containerd/runwasi.git \
--bin containerd-shim-wasmtime-v1 \
--root /out \
containerd-shim-wasmtime
FROM scratch
COPY --from=build /out/bin /
EOF
To run the wasm docker image:
docker run --runtime=io.containerd.wasmtime.v1 --platform=wasi/wasm kcandidate/hello-wasm-world:v1
Running python hello-wasm-world.py:
Elapsed time for range sum: 0.001385 seconds
Pystone(1.1) time for 50000 passes = 0.124373
This machine benchmarks at 402016 pystones/second
Running wasmer run hello-wasm-world.wasm:
Elapsed time for range sum: 0.000775 seconds
Pystone(1.1) time for 50000 passes = 0.227372
This machine benchmarks at 219904 pystones/second
Running docker run --runtime=io.con tainerd.wasmtime.v1 --platform=wasi/wasm32 kcandidate/hello-wasm-world:v1:
Elapsed time for range sum: 0.000921 seconds
Pystone(1.1) time for 50000 passes = 0.238069
This machine benchmarks at 210023 pystones/second