I want to transfer efficiently my images to the production machines.
My Dockerfile with android sdk
built with docker build -t docker_layers .
I want to push it to the registry
It will push 3 layers:
the first is the android tgz pushed into /opt
the second is the extracted sdk into the /opt directory
the last one is the deletion of the tgz
This is really not optimal, the first and third layer are useless. Worst, they use space on your filesystems, on your network when you pull/push, …
It is a good practice to merge common operations into one single docker command.
For this example it would be:
In that way you only have one layer commited, and it contains only the needed android sdk extracted.
There is one drawback with this particular approach: You lose the capacities of the ADD syntax.
If you build the first Dockerfile, you will notice that each time the tgz is downloaded. Docker then verifies that the file hasn’t changed and if not, uses the cached image.
That means that if the tgz changes, there will be a new build, whereas with the one-line approach, it will not, because docker will just verify that the command hasn’t changed, not the content downloaded by the wget.
The general approach is :
Prepare what you want to do
Clean everything that is not necessary
The example with the android sdk works for all archives that must be extracted.
Below are templates from our experience that we think doesn’t alter readability while reducing the layers and the size of the docker image.
An other very important thing is to keep your Dockerfile readable
This example is, in my opinion, a bad one.
Yes, it’s only one docker RUN command
If you know other tips to reduce the number of layers without losing readability, feel free to post a comment, I’ll gladly add them here.