Problems with pushing to ECR

[teofiln]

As described here richfitz/stevedore#61 and here richfitz/stevedore#63

Maybe call docker externally with a system command while the stevedore issue gets resolved.

[DyfanJones]

Not sure if you're aware, but you can push docker images to aws ecr by using smdocker. It is based off the python cli sm-docker.

A little background information around smdocker. The desired directory containing the dockerfile is zipped up and and pushed to AWS Codebuild to build the docker image. It is then outputs the results to AWS ECR.

The pros with this approach, is that users doesn't require docker to setup on their local computer. Here is an AWS Blog in how it works: https://aws.amazon.com/blogs/machine-learning/using-the-amazon-sagemaker-studio-image-build-cli-to-build-container-images-from-your-studio-notebooks/

[teofiln]

Hi @DyfanJones! Thanks for this! I'll take a look and be in touch.

[teofiln]

Seems like the error Error in log(bytes, 1000) : non-numeric argument to mathematical function that happens sometimes when pushing the image to ECR is down to a function in stevedore that formats the push progress output. I bypased this by removing the calls to the formatting function in my branch of stevedore (here: https://github.com/teofiln/stevedore/blob/38034eb9e4d1a23a06f6e7353dbcdf86edcf09ac/R/docker_client_support.R#L121).

[teofiln]

I'd like to keep the functionality to be able to test the function locally before pushing to ECR. I am not sure how something like this would work if the image is built by smdocker. So, at least for now, we'll keep the system dependency for docker. Maybe later we can add the option to build the image with smdocker, while maintaining the local docker option as well.