NX monorepo: Deploy and Run Node microservices locally with Docker and Kubernetes

[creotip]

Step by step guide I wrote on NX, Docker and Kubernetes

https://creotip.io/posts/nx-monorepo-running-microservices-locally-with-docker-kubernetes

[fider]

1. What about dependencies? I don't want to have extra 920MB of dependencies installed in image A, B, C (...and 90 others) only because one of developers required some dependencies for service X. Any suggestion how to bypass this? How to install dependencies required only by service A when deploying service A?

# dependencies that express needs
RUN npm install reflect-metadata tslib rxjs express

Imo. not production ready solution. Any suggestions on how to avoid this? implicitDependencies in project.json or dependencies in package.json approach?

[juristr]

Since Nx has an internal project graph (which also happens to include npm packages), it perfectly knows which packages are used by a given project. As such you can generate a package.json during the build which can be used for installing only those packages required by the given project in the monorepo.

Many of the built-in Nx executors have this already built-in. Look for the generatePackageJson flag or similar.

Also, there's a potentially helpful discussion here how to write your own executor to do exactly that:
nrwl/nx-labs#31

[BBlackwo]

+1 for for the 2nd issue you mentioned @fider. generatePackageJson doesn't pick up some dependencies such as reflect-metadata (or mysql when using typeorm). A different workaround is to explicitly import those packages in your main.ts and then they will be added to the generated package.json:

import 'mysql';
import 'reflect-metadata';

BTW the official Nx and Node Microservices guide uses an explicit install in the Dockerfile, same as @creotip:

# dependencies that nestjs needs
RUN npm install reflect-metadata tslib rxjs @nestjs/platform-express

There's also this related issue #1777 (package.json per app)