Docker is a file based system and for each Pull request, it only pulls out the files which are changed. For example, if suppose you have 1 GB data in a file in a docker image. Now, you added 500MB of data to it. Then, in case of docker pull, it will only pull the changes, i.e the Delta part between the 2 files. So, you are safe and it won't pull all the things separately.

Although, while making a DockerFile, or docker conf file, you should be very careful, as all the lines in a Docker file is stored as a layer in the system. If suppose you have 10 layers in your Docker file, and you are changing the 5th layer, then all the layers after the 5th layer will again be pulled. That is the only catch using Docker. Rest, it will always pull the Delta of changes for each pull request.

If you want to save/load tar files of docker images, there's no option to export a partial image. You can send over the full image, move your data to be an external volume that isn't transfered this way, or you can use a docker registry.

The latter is relatively easy to implement, docker includes an image where you can run your own private registry. Pushes and pulls to a docker registry will only send the changed layers, so you can make use of layer caching and structure your Dockerfiles to minimize the number of changed layers.

Ok, I've built the tool to create diff (in top layers) of docker images' versions (layer by layer) as a tarball and inflate the original image later.

Note. Works only for changes in top layers.

4-step process:

1. docker inspect -> make json with old layers' hashes as a json file
2. Prepare diff based on new image and hashes of old (existing) layers
3. Transfer diff to target machine
4. Inflate target image's tar based on diff and old image