Use a docker cache mount to speed up rebuilding docker images after a dependency change

rubyonrails-core
1

Adding, removing, or updating gems currently have bad cache misses for the default docker image produced with rails new . This particularly affects rails apps with many native dependencies that must be rebuilt. By using a docker mount cache the build process can reuse the work that was previously done to fetch and install dependencies.

On a freshly created rails app using postgres, adding sidekiq (after initially building the docker image without the sidekiq gem), a mount cache halves the build time.

# Without a mount cache adding sidekiq
> bundle add sidekiq && time docker build .
docker build .  0.23s user 0.21s system 1% cpu 35.999 total

# With a mount cache adding sidekiq
> bundle add sidekiq && time docker build .
docker build .  0.17s user 0.13s system 1% cpu 18.277 total

The downside to this change is the Dockerfile becomes more complicated

RUN --mount=type=cache,target=/tmp/bundle-cache \
    cp -r /tmp/bundle-cache/* "${BUNDLE_PATH}/" 2>/dev/null || true && \
    bundle install && \
    cp -r "${BUNDLE_PATH}"/* /tmp/bundle-cache/ && \
    rm -rf ~/.bundle/ "${BUNDLE_PATH}"/ruby/*/cache "${BUNDLE_PATH}"/ruby/*/bundler/gems/*/.git && \
    # -j 1 disable parallel compilation to avoid a QEMU bug: https://github.com/rails/bootsnap/issues/495
    bundle exec bootsnap precompile -j 1 --gemfile

vs

RUN bundle install && \
    rm -rf ~/.bundle/ "${BUNDLE_PATH}"/ruby/*/cache "${BUNDLE_PATH}"/ruby/*/bundler/gems/*/.git && \
    # -j 1 disable parallel compilation to avoid a QEMU bug: https://github.com/rails/bootsnap/issues/495
    bundle exec bootsnap precompile -j 1 --gemfile

Is there any appetite for a pull request here?