Is it possible to create an AWS AMI from a Docker image?

Question

I have a Docker image in my AWS ECR which I want to convert to an AWS AMI for creating an instance with the environment.

Reason for trying to use an AMI over a Docker container

I had to run computationally intensive ML tasks in my Docker container inside an m4.xlarge instance on some text files. However, with Docker, the CPU utilization of the instance is reaching 100% and the script is getting killed pre-maturely.

However, it is running fine in the same instance from outside of Docker.

---

So, I want to move to AMI's from Docker, however I also want to leverage Docker for the environment creation and the speed of testing which it enables in my local.

So, is it possible to create an AMI out of the Docker image which I test and deploy to my ECS registry?

Answer 1

There is no direct way to create an AWS AMIs out of docker image although there is an opensource tool which can create/build AWS AMIs out of a docker file but only supported instructions are ENV, RUN, COPY and ADD, other instructions will just be ignored.

Open-source Tool for Building AWS AMI from Docker file

https://github.com/mickep76/docker-build-ami

https://pypi.python.org/pypi/docker-build-ami/0.1.3

So if you are really interested in trying out go ahead with creating a proper docker file from your docker image. For that you can refer following links

Is it possible to extract the Dockerfile from a docker container

How to generate a Dockerfile from an image?

Answer 2

It would seem that there is a problem or misconfiguration with the docker image or docker. You mention the instance runs the same script outside of docker without a problem? Then better create an AMI of the instance with the running script and you don't need docker at all. Having a script running within docker could at most mean the docker configuration limits CPU, but not that the host CPU goes to 100% and kills the script. You should try running a docker container with the same image, but on a sleep10000 run, so you can go within the container, run the script of interest and see what is failing.