I have recently discovered Docker's new multi-stage build feature. This has been a great help in answering my question of how to reduce the size of my haskell images for deploying, as the main haskell image on Docker Hub is over 1GB before you start adding things. This also means you don't have to maintain two Dockerfiles.
With multi stage builds we can use the
haskell docker image to build our distributable executable, then copy this to a smaller image (in this case
Here's an example Dockerfile for a project called
FROM haskell as build-env WORKDIR /opt/server RUN cabal update COPY haskell-example.cabal . RUN cabal install --only-dependencies -j4 ADD . . RUN cabal install FROM phusion/baseimage WORKDIR /opt/server COPY --from=build-env /opt/server/dist . CMD ./build/haskell-example/haskell-example
Before, using just the haskell image: 1.87GB.
After, using multi-stage with baseimage: 249MB.
There's just one problem with this method - the intermediate images are all marked as dangling. This means that if you do a
docker system prune and rebuild, you will have install all your dependencies again, which takes a lot of time in the haskell world.
The workaround for this is to build an image using the
build-env stage. This is done by specifying the target when building.
docker build -t haskell-example-dev --target build-env .
This keeps the image in the cache so it won't be
So we've shrunk our deployable image using
phusion/baseimage, can we go further? The next step would be to use something like alpine as the production image, a mere 4.14MB. That's what I'll talk about next time :)