When we were trying to create a docker image for our Node JS based application, we chose to use the official Node docker image (~700MB). On top of that we need to add the node modules, business logic, etc and so on. The final image size was staggering (~1.2GB). It was not what we wanted. Secondly, the average build time to do NPM install and run a grunt task totally took 15 minutes for every build. I am not even talking about the pain of configuring this for different CI/CD pipelines and environments.
The initial docker file was looking something like this:
FROM node:6.10.1-alpine
ENV APP_ENV $APP_ENV
ENV BINPATH /usr/bin
COPY wkhtmltox/bin/wkhtmltopdf /usr/local/bin/
RUN mkdir -p /usr/src/app
WORKDIR /usr/src/app/
# Install app dependencies
COPY package.json /usr/src/app/
#COPY start.sh /usr/src/app/
RUN npm install --production
RUN npm install -g grunt-cli@1.2.0
# Bundle app source
COPY . /usr/src/app
# Install pm2 so we can run our app
RUN npm i -g pm2
# run grunt, cd to dist, node server.js
RUN grunt --env=prod
#COPY start.sh dist
WORKDIR dist
ADD start.sh /start.sh
RUN chmod 755 /start.sh
CMD ["/start.sh"]
- Docker Image Size
- Image build Time
- Configure the image building process for different environment pipeline
I first decided to upgrade to Node:6 because of the LTS (Long Term Support), next I wanted to use a slim version of offcial node docker image. I didn't choose the alpine version for the reason there is no official support for node-sass. There are plethora of docker images available with sass-alpine-node version. But I strongly recommend not to use it for security reasons.
FROM node:6-slim
Secondly I want to build a base web image specific to my project with all the node_modules dependencies installed. Whenever we add a new module to npm package, we need to build the base image once which is not going to happen super frequently.
The next thing, I did after updating to node 6 was to make use of in built node capabilities. You can read more about them in the official npmjs website more.
Removes the duplicate refrences of node modules
npm dedupe
Removes the unreferenced node modules
npm prune
Create a shrinkwrap to lock down the node module versions
npm shrinkwrap
This has reduced the number of modules get installed and the size of the node_modules directory. The final base image docker file looked like below.
FROM node:6-slim
ENV APP_ENV $APP_ENV
ENV BINPATH /usr/bin
COPY wkhtmltox/bin/wkhtmltopdf /usr/local/bin/
RUN mkdir -p /usr/src/app
COPY package.json /usr/src/app/
WORKDIR /usr/src/app/
RUN npm install --production
Then I built a base image like below:
docker build -t base-web-image:<version-number> .
After this I removed the executing the grunt task out the docker image build process and moved it to the build agent machine.
grunt --env=prod
As a final step, whenever a business logic or new code is pushed a new image needs to be built as a final application. I refered the image built earlier and started to refer it in the second image.
FROM base-web-image
RUN mkdir -p /usr/src/app
COPY dist /usr/src/app
WORKDIR /usr/src/app
CMD ["pm2 start"]
- Final image size: ~400MB (~1.2 GB earlier)
- Build time: 10 seconds (15 minutes earlier)
Note: All the image sizes mentioned are uncompressed version sizes. The actualy size uploaded to docker registries will be compressed and will smaller than the sizes mentioned.
Next Steps: Need to explore options to reduce the size by exploring the official wheezy/alpine versions and install node by my self and test.
Nice Sundar