An experimental NodeJS+Docker workflow

Recently, I’ve been trying hard to integrate Docker into my development workflow. On a server where I deploy tiny tools internally I’ve decided that I don’t want to have to manage 10 different environments but simply deploy cronjob-scripts, services and more as Docker images.

So problems I wanted to address are:

  • I want to have a single Dockerfile for use during development as well as production. This means that sharing the node_modules folder through volumes has to be avoided (this will soooo break on OSX 😉).
  • During development (esp. for small service with a web-interface) I want to be able to have some kind of file-monitoring going on so that I don’t have to restart the container every time I change something.

The result of this experiment has been this little Dockerfile:

FROM node:4

# Prepare non-root user and folders
RUN useradd --system --user-group --create-home app && \
    mkdir /app && chown app:app /app
RUN npm install -g nodemon

# Install dependency outside of the app volume
COPY package.json /opt/
RUN cd /opt && npm install
ENV NODE_PATH=/opt/node_modules

VOLUME ["/app"]
USER app
WORKDIR /app
EXPOSE 8080
ENTRYPOINT ["/app/entrypoint.sh"]

This will install all the dependencies the project has inside a folder outside the primary working directory and therefore avoid linking issues esp. with npm3.

The entrypoint.sh is simply whatever I need to use this in production. To make switching between dev and prod easier, I try to keep all the dev-configuration inside the docker-compose.yml file which mostly just mounts the project’s folder into /app and overrides the entrypoint (or for different setups simply sets command parameters):

version: '2'
services:
  app:
    build: .
    volumes:
      - .:/app
    ports:
      - 8080:8080
    entrypoint: nodemon -L main.js

For more complex systems it’s trivial to add dependencies like PostgreSQL or Redis here.

The big downside so far is the -L flag you can see in the entrypoint specification. Without it nodemon won’t notice any changes within the volume but with it it’s doing quite a bit of busy-work which wastes power. I haven’t found a good solution for this yet. If you have, please let me know 😃

For smaller things like scripts and cronjob this is a non-issue as I can execute them directly with docker run 🙂

I’ve also created a full example on this gist for your cloning pleasure 😉

Not a primary goal so far was keeping the container size down as the machine has enough disk space available. That being said, I’ll probably try to find a good balance between the slim and non-slim base images in the future. For other environments I’m also try to stick with Alpine where ever it makes sense 😊

comments powered by Disqus