dockerfile-from-image
Reverse-engineers a Dockerfile from a Docker image.
Similar to how the docker history command works, the dockerfile-from-image
script is able to re-create the Dockerfile (approximately) that
was used to generate an image using the metadata that Docker stores alongside
each image layer.
Usage
The Ruby dockerfile-from-image script is itself packaged as a Docker image so it can easily be executed with the Docker run command:
docker run -v /var/run/docker.sock:/var/run/docker.sock \
centurylink/dockerfile-from-image <IMAGE_TAG_OR_ID>
The <IMAGE_TAG_OR_ID> parameter can be either an image tag (e.g. ruby) or
an image ID (either the truncated form or the complete image ID).
Since the script interacts with the Docker API in order to query the metadata
for the various image layers it needs access to the Docker API socket. The
-v flag shown above makes the Docker socket available inside the container
running the script.
Note that the script only works against images that exist in your local image
repository (the stuff you see when you type docker images). If you want to
generate a Dockerfile for an image that doesn't exist in your local repo
you'll first need to docker pull it.
Example
Here's an example that shows the official Docker ruby image being pulled and the Dockerfile for that image being generated.
$ docker pull ruby
Pulling repository ruby
$ docker run --rm -v /run/docker.sock:/run/docker.sock centurylink/dockerfile-from-image
Usage: dockerfile-from-image.rb [options] <image_id>
-f, --full-tree Generate Dockerfile for all parent layers
-h, --help Show this message
$ docker run --rm -v /run/docker.sock:/run/docker.sock centurylink/dockerfile-from-image ruby
FROM buildpack-deps:latest
RUN useradd -g users user
RUN apt-get update && apt-get install -y bison procps
RUN apt-get update && apt-get install -y ruby
ADD dir:03090a5fdc5feb8b4f1d6a69214c37b5f6d653f5185cddb6bf7fd71e6ded561c in /usr/src/ruby
WORKDIR /usr/src/ruby
RUN chown -R user:users .
USER user
RUN autoconf && ./configure --disable-install-doc
RUN make -j"$(nproc)"
RUN make check
USER root
RUN apt-get purge -y ruby
RUN make install
RUN echo 'gem: --no-rdoc --no-ri' >> /.gemrc
RUN gem install bundler
ONBUILD ADD . /usr/src/app
ONBUILD WORKDIR /usr/src/app
ONBUILD RUN [ ! -e Gemfile ] || bundle install --system
Run it as local command
$ docker pull centurylink/dockerfile-from-image
$ alias dfimage="docker run --rm -v /var/run/docker.sock:/var/run/docker.sock centurylink/dockerfile-from-image"
$ dfimage --help
Usage: dockerfile-from-image.rb [options] <image_id>
-f, --full-tree Generate Dockerfile for all parent layers
-h, --help Show this message
$ dfimage ruby
How Does It Work?
When an image is constructed from a Dockerfile, each instruction in the
Dockerfile results in a new layer. You can see all of the image layers by
using the docker images command with the (soon-to-deprecated) --tree flag.
$ docker images --tree
Warning: '--tree' is deprecated, it will be removed soon. See usage.
└─511136ea3c5a Virtual Size: 0 B Tags: scratch:latest
└─1e8abad02296 Virtual Size: 121.8 MB
└─f106b5d7508a Virtual Size: 121.8 MB
└─0ae4b97648db Virtual Size: 690.2 MB
└─a2df34bb17f4 Virtual Size: 808.3 MB Tags: buildpack-deps:latest
└─86258af941f7 Virtual Size: 808.6 MB
└─1dc22fbdefef Virtual Size: 846.7 MB
└─00227c86ea87 Virtual Size: 863.7 MB
└─564e6df9f1e2 Virtual Size: 1.009 GB
└─55a2d383d743 Virtual Size: 1.009 GB
└─367e535883e4 Virtual Size: 1.154 GB
└─a47bb557ed2a Virtual Size: 1.154 GB
└─0d4496202bc0 Virtual Size: 1.157 GB
└─5db44b586412 Virtual Size: 1.446 GB
└─bef6f00c8d6d Virtual Size: 1.451 GB
└─5f9bee597a47 Virtual Size: 1.451 GB
└─bb98b84e0658 Virtual Size: 1.452 GB
└─6556c531b6c1 Virtual Size: 1.552 GB
└─569e14fd7575 Virtual Size: 1.552 GB
└─fc3a205ba3de Virtual Size: 1.555 GB
└─5fd3b530d269 Virtual Size: 1.555 GB
└─6bdb3289ca8b Virtual Size: 1.555 GB
└─011aa33ba92b Virtual Size: 1.555 GB Tags: ruby:2, ruby:2.1, ruby:2.1.1, ruby:latest
Each one of these layers is the result of executing an instruction in a
Dockerfile. In fact, if you do a docker inspect on any one of these layers
you can see the instruction that was used to generate that layer.
$ docker inspect 011aa33ba92b
[{
. . .
"ContainerConfig": {
"Cmd": [
"/bin/sh",
"-c",
"#(nop) ONBUILD RUN [ ! -e Gemfile ] || bundle install --system"
],
. . .
}]
The output above has been truncated, but nested within the ContainerConfig
data you'll find the Dockerfile command that generated this layer (in this case
it was an ONBUILD instruction).
The dockerfile-from-image script works by simply walking backward through the
layer tree and collecting the commands stored with each layer. When the script
reaches the first tagged layer (or the root of the tree) it stops and displays
the (reversed) list of commands. If you want to generate the commands going
all the way back to the root image layer you can use the -f flag to walk the
entire tree.
Limitations
As the dockerfile-from-image script walks the list of layers contained in the
image it stops when it reaches the first tagged layer. It is assumed that a layer
which has been tagged represents a distinct image with its own Dockerfile so the
script will output a FROM directive with the tag name.
In the example above, the ruby image contained a layer in the local image repository which had been tagged with buildpack-deps (though it wasn't shown in the example, this likely means that buildpack-deps:latest was also pulled at some point). If the buildpack-deps layer had not been tagged, the dockerfile-from-image script would have continued outputing Dockerfile directives until it reached the root layer.
Also note that the output generated by the script won't match exactly the
original Dockerfile if either the COPY or ADD directives (like the
example above) are used. Since we no longer have access to the build context
that was present when the original docker build command was executed all we
can see is that some directory or file was copied to the image's filesystem
(you'll see the file/directory checksum and the destination it was copied to).