Docker Build Cacher

This tool is intended to speedup multi-stage Dockerfile build times by caching the results of each of the stages separately.

Why?

Multi-stage docker file builds are great, but they still miss a key feature: It is not possible to carry from one build to another the statically generated cache files once the source file in your project change. Here's an example that illustrates the issue:

Imagine you create a generic Dockerfile for building node projects

FROM nodejs

RUN apt-get install nodejs yarn

WORKDIR /app

# Whenever this image is used execute these triggers
ONBUILD ADD package.json yarn.lock .
ONBUILD RUN yarn
ONBUILD RUN yarn run dist

And then you call

docker build -t nodejs-build .

So now you can use the nodejs-build image in other builds, like this:

# Automatically build yarn dependencies
FROM nodejs-build as nodedeps

# Build the final container image
FROM scratch

# Copy the generated app.js from yarn run dist
COPY --from=nodedeps /app/app.js .
...

So far so good, we have build a pretty lean docker image that discards all the node_modules folder and only keeps the final artifact. For example a bundled reactjs application.

It's also very fast to build! Since each of the steps in the Dockerfile are cached, as long as none of the files changed.

But that's also where the problem is: Whenever package.json or yarn.lock files change, docker will trash all the files in node_modules and all the cached yarn packages and will start from scratch downloading, linking and building every single dependency.

That's far from ideal. What if we could do a change in the process so that changes to those files do not bust the yarn cache? It turns out that we can!

Enter docker-build-cacher

This utility overcomes the problem by providing a way to build the docker file and then cache the intermediate stages. On subsequent builds, it will make sure that the static cache files generated during previous builds will also be present.

The effect it has should be obvious: your builds will be consistently fast, at the cost of more disk space.

Installation

There are binaries provided for linux-x86_64 and MacOS, check the releases page for downloads.

How It Works

This works by parsing the Dockerfile and extracting the COPY or ADD instructions nested inside ONBUILD for each of the stages found in the file.

It will compare the source files present in such COPY or ADD instructions to check for changes. If it can detect changes, it rewrites your Dockerfile on the fly so that the FROM directives in each of the stages use the locally cached images instead of the original base image.

The effect this FROM swap has, is that disk state for the image is preserved between builds.

Usage

docker-build-cacher requires the following environment variables to be present in order to correctly build your Dockerfile:

• APP_NAME: The name for application you are trying to build. Usually this is just the folder name you are in.
• GIT_BRANCH: The name of the git branch you are building. Used to "namespace" cache results
• DOCKER_TAG: It will docker build -t $DOCKER_TAG . at some point. Let it know the image tag you want at the end.

This utility has two modes, Build and Cache. Both modes should be invoked for the cache to work:

# APP_NAME ispassed as argument in the build process, you can use it as an env var in your Dockerfile
export APP_NAME=fancyapp

# GIT_BRANCH is used as part of the named for the resulting cached image
export GIT_BRANCH=master

# DOCKER_TAG corresponds to the -t argument in docker build, that will be the resulting image name
export DOCKER_TAG=fancyapp:latest

docker-build-cacher build # This will build the docker file
docker-build-cacher cache # This will cache each of the stage results separately

Additionally, docker-build-cacher accepts the DOCKERFILE env variable in case the file is not present in the current directory:

DOCKERFILE=buildfiles/Dockerfile docker-build-cacher build

At the end of the process you can call docker images and see that it has created fancyapp:latest, and if you are using multi-stage builds, it should have created an image tag for each of the stages in your Dockerfile

Fallback Cache Keys

As mentioned before the GIT_BRANCH env variable is used as part of the name for the generated cached image, this means that the generated cache is scope to that name. This is done so you can keep different caches where you can experiment with widly different requirements and libraries in the dockerfile.

This has the unfortunate side effect that building other branches will require building the cache from scratch. In order to solve this you can use the FALLBACK_BRANCH environment variable like this:

export APP_NAME=fancyapp
export GIT_BRANCH=my-feature
export FALLBACK_BRANCH=master
export DOCKER_TAG=fancyapp:latest

docker-build-cacher build
docker-build-cacher cache

The above will make the cached image for the my-feature branch to be based on the one from the master branch.

Caching Intermediate Images

In some circumstances, you may want to execute additional instructions after including the base builder image. For instance, building an executable or bundle using all the dependencies already downloaded:

# Automatically build haskell stack dependencies
FROM haskell-stack as builder

COPY . .
RUN stack install

# Build the final container image
FROM scratch

COPY --from=builder /root/.local/bin/my-app

This very typical example has a shortcoming now, each time we do COPY . . we are also invalidating the compiling artifacts created in stack install, that is, we are losing the benefits of incremental compilation.

If you want to keep incremental compilation, or any files generated in between the builder image and the final FROM, you can label the intermediate image so that docker-build-cacher will include that into the cached artifacts:

# Automatically build haskell stack dependencies
FROM haskell-stack as builder

# Instructs the cacher to also copy the files generated in this stage
LABEL cache_instructions=cache

COPY . .
RUN stack install

# Build the final container image
FROM scratch

COPY --from=builder /root/.local/bin/my-app

Warning:

The files copied in COPY . . will also be cached! This not only increases the cache size, but also has a potentially dangerous inconvenient:

Any files you delete from one build to the other will be restored again by the cacher. For example, if you delete one file in your source tree because you don't use it anymore or you did a refactoring, it will pop up again in the build!

This may be a problem for compilers or build tools that scan all the files in the folder, like the Go compiler. If you are certain that keeping old files around is not a problem, then it is safe to use this feature. The Haskell compiler, for instance, does not care at all about extra cruft in the folder.

Passing extra arguments to docker build

It is possible to pass extra arguments and flags to the docker build step by providing the environment variable DOCKER_BUILD_OPTIONS as shown below:

DOCKER_BUILD_OPTIONS="--build-arg foo=bar --quiet" docker-build-cacher build

Building from source

Dependencies:

• Haskell stack

Install the stack tool from the link above. Then cd to the root folder of this repo and execute:

stack setup
stack install

If it is the first time, it will take a lot of time. Don't worry, it's only once you need to pay this price.