Yarn cache compatible Dockerfile

README.md

Yarn Dockerfile

An yarn cache compatible Dockerfile, for building node.js images faster.

Usage

Clone this gist into your project root, and add it to your source control. Change the image service-name:latest tag to your project name in the Dockerfile and build.sh files. Then, always build your image using the build.sh script.

$ chmod +x ./build.sh
$ ./build.sh

build.sh

#!/bin/bash

if [ ! -f .yarn-cache.tgz ]; then
  echo "+ build: Init empty .yarn-cache.tgz"
  tar cvzf .yarn-cache.tgz --files-from /dev/null
fi

docker build -t service-name:latest .

docker run \
  --rm \
  --entrypoint cat \
  service-name:latest \
  /tmp/yarn.lock > /tmp/yarn.lock

if ! diff -q yarn.lock /tmp/yarn.lock > /dev/null  2>&1; then
  echo "+ build: Saving Yarn cache"

  docker run \
    --rm \
    --entrypoint tar \
    service-name:latest \
    czf - /root/.yarn-cache/ > .yarn-cache.tgz

  echo "+ build: Saving yarn.lock"

  cp /tmp/yarn.lock yarn.lock
fi

Dockerfile

FROM alpine

RUN apk add --update --no-cache nodejs
RUN npm i -g yarn

ADD package.json yarn.lock /tmp/
ADD .yarn-cache.tgz /

RUN cd /tmp && yarn
RUN mkdir -p /service && cd /service && ln -s /tmp/node_modules

COPY . /service
WORKDIR /service

ENV FORCE_COLOR=1

ENTRYPOINT ["npm"]
CMD ["start"]

In case someone arrives here wondering why yarn install doesn't even cache locally, check out your COPY commands beforehand, and split them in two:

WORKDIR /app
COPY package.json yarn.lock .yarnrc ./
RUN yarn --pure-lockfile --non-interactive
COPY . ./

If you just do COPY . ./, it's very very likely something has changed in your project, and caching will never work.

In case someone arrives here wondering why yarn install doesn't even cache locally, check out your COPY commands beforehand, and split them in two:

WORKDIR /app
COPY package.json yarn.lock .yarnrc ./
RUN yarn --pure-lockfile --non-interactive
COPY . ./

If you just do COPY . ./, it's very very likely something has changed in your project, and caching will never work.

If you are also wondering why your container doesn't work after building on OSX, its because yarn has issues symlinking on OSX/BSD

@etgrieco The provided code works. On the initial container start up the dependencies need to be downloaded and then are stored into the volume. When the container it is stoped and started again, dependencies does not need to re-installed but yarn take some time to "Link the dependencies", and this happens on every container restart. Any idea about that ? It take a good amount of time in my end (~20 seconds to link the dependencies).

@etgrieco The provided code works. On the initial container start up the dependencies need to be downloaded and then are stored into the volume. When the container it is stoped and started again, dependencies does not need to re-installed but yarn take some time to "Link the dependencies", and this happens on every container restart. Any idea about that ? It take a good amount of time in my end (~20 seconds to link the dependencies).

I haven't used this docker-based workflow in a while, but hopefully others who ran into your issue can chime in!

@etgrieco Ah, okay. I am developing a Next.js app, and its dependency installation is quite annoying. Without such a "trick", when I change something in package.json, it takes over 150 + seconds just for the yarn install step when a rebuild the Docker Image. So, any trick/optimization technique would be greatly appreciated!