Docker and AWS

[dan9thsense]

I tried following the documentation for using AWS with docker. The AWS instance is a Basic Deep Learning Ubuntu build on a p2.xlarge with 65 GB storage, as specified in the docs. The dockerfile is the same as I use for submissions for a score. I almost got it to run, but it has an error:

'Academy' started successfully!
Unity Academy name: Academy
        Number of Brains: 1
        Number of Training Brains : 1
INFO:gym_unity:1 agents within environment.
: cannot connect to X server 

This is what I used to start the docker image:

docker run --runtime=nvidia -v "$PWD"/test_submission:/aaio/test submission_scripts python /aaio/test/testDocker.py

[beyretb]

I ran your latest submission following the instructions from the documentation and managed to run your command just fine.

Are you using the recommended deep learning AMI? It should work with this one (you can remove the folders present in the home directory if you're not using them -especially anaconda- as they take a lot of space).

[dan9thsense]

That's odd. I am using the suggested Deep Learning AMI Ubuntu, p2.xlarge, as specified in the documentation: Deep Learning Base AMI (Ubuntu 16.04) Version 19.2 - ami-0dda3a10b93f38b21

I would need to get more permissions (for 8 vCPUs) to run the instance you linked to. It seems that the issue is with opening a graphical window (cannot connect to X server).

[dan9thsense]

I uploaded the docker image that I submitted and you tested and, sure enough, it runs on the AWS machine. There must be something wrong with the docker image that I created on the AWS machine. I'll investigate further.

[dan9thsense]

Works fine now. Never figured it out, just rebuilt the docker image from scratch and it works.

[beyretb]

glad you managed to fix the issue!