Using NODE_ENV and building for multi-arch in Yarn 3

[maggie44]

I am struggling a little to get my head around an appropriate workflow for Yarn 3 in official Node docker images (using': node:16.13.2-alpine3.15).

It seems when I add the NODE_ENV=production environment flag to the Dockerfile, and do a build it is ignored and it installs all the development packages. This is the case both with nodeLinker: node-modules and without. Is this expected?

Here is a process to test. When I use this process with yarn v1 without running yarn set version stable it only installs dependencies, not development dependencies but the below always installs both.

docker run -it node:16.13.2-alpine3.15 sh
yarn set version stable
yarn --version
	3.2.0
mkdir test
cd test
apk add git
yarn init
yarn add -D typescript
yarn add axios
NODE_ENV=production yarn install
ls .yarn/cache
	axios-npm-0.26.1-a6641ce4e3-d9eb58ff4b.zip
	typescript-npm-4.6.3-1493ebc82b-255bb26c8c.zip
	follow-redirects-npm-1.14.9-522f191631-f5982e0eb4.zip
	typescript-patch-30b732d1e2-6bf45caf84.zip

echo 'nodeLinker: node-modules' >> .yarnrc.yml
NODE_ENV=production yarn install
ls ./node_modules
	axios             
	follow-redirects  
	typescript

A separate question. I am building for multi-arch images, and wondering what implications this may have for 'Zero Install'. Will a Yarn install on my local system (Mac M1) have implications for any node installs that may be architecture specific? Should we always do the Yarn install to generate the cache on an architecture equivalent to the one we are deploying to as a rule of thumb? Maybe I missed something, but couldn't see anything documented on it.

There are some example docker builds here, but wow, it's really convoluted to get yarn and docker to work like this. Has anyone come across a better way?