build context for docker image very large

I have created a couple different directories on my host machine as I try to learn about Docker just to keep my dockerfiles organized. My Dockerfile I just ran looks like this:

FROM crystal/centos
MAINTAINER crystal

ADD ./rpms/test.rpm ./rpms/ 
RUN yum -y --nogpgcheck localinstall /rpms/test.rpm 

My actual rpm is only 1 GB. But when I try to do sudo docker build -t="crystal/test" ., I get sending build context to Docker daemon 3.5 GB. Is there something else that I'm unaware of as you continue to build Docker images? Is my memory accumulating as I build more images in my other directories on my host machine?

The Docker client sends the entire "build context" to the Docker daemon. That build context (by default) is the entire directory the Dockerfile is in (so, the entire rpms tree).

You can setup a .dockerignore file to get Docker to ignore some files. You might want to experiment with it.

Alternatively, you can move your rpms folder one directory level above your Dockerfile, and only symlink test.rpm into the Dockerfile's directory.

You’ll often want to add the .git folder to the .dockerignore which was the cause of a 150MB -> 5GB difference for some users in the comments here.

Update 2019

Starting from Docker v18.06 there is an option to use a new image builder called Build Kit.

It's pre-bundled with the Docker, no need to install anything. It's backward compatible with the Dockerfile syntax, no need to change the Dockerfile.

Legacy Docker Build vs New Docker BuildKit

Here is an example of building an image with a huge unused file in the build directory:

Legacy Docker Build:

$ time docker image build --no-cache .
Sending build context to Docker daemon  4.315GB
[...]
Successfully built c9ec5d33e12e

real    0m51.035s
user    0m7.189s
sys 0m10.712s

New Docker BuildKit:

$ time DOCKER_BUILDKIT=1 docker image build --no-cache .
[+] Building 0.1s (5/5) FINISHED                                                
 => [internal] load build definition from Dockerfile                       0.0s
 => => transferring dockerfile: 37B                                        0.0s
 => [internal] load .dockerignore                                          0.0s
 => => transferring context: 2B                                            0.0s
[...]
 => => writing image sha256:ba5bca3a525ac97573b2e1d3cb936ad50cf8129eedfa9  0.0s

real    0m0.166s
user    0m0.034s
sys 0m0.026s

The only change is the DOCKER_BUILDKIT=1 environment variable, the difference in time is huge.

.dockerignore File

Please note, that the .dockerignore file is still valid and useful. Some Dockerfile commands like COPY . . will still take into account the .dockerignore rules. But the side files in the build directory (not referenced in the Dockerfile) are not getting copied anymore as a "build context" by the BuildKit.

I fixed it by moving my Dockerfile and docker-compose.yml into a subfolder and it worked great. Apparently docker sends the current folder to the daemon and my folder was 9 gigs.

If you have a .dockerignore file and build context is still large, you can check what is being sent to the docker build context using The Silver Searcher:

ag --path-to-ignore .dockerignore --files-with-matches

Note that some ** patterns might not work properly.

See this Github issue for additional comments: https://github.com/moby/moby/issues/16056