Add publish_container.py

[ax3l]

Generalize our container build and publish script to also build the ML container #302.

[EZoni]

How much of this code is similar to the one in publish_dashboard.py?

I would be in favor of reusing as much as code as possible and avoid duplication as much as possible.

Here's the diff:

0a1,5
> #!/usr/bin/env python
> """
> Build the ML training container (with CUDA support) for
> Perlmutter (NERSC) and publish it to registry.nersc.gov
> """
28c33
<     command = "docker build --platform linux/amd64 -t gui -f dashboard/Dockerfile ."
---
>     command = "docker build --platform linux/amd64 -t gui -f ml/Dockerfile ."
35,37c40,42
<         "docker tag gui:latest registry.nersc.gov/m558/superfacility/gui:latest",
<         "docker tag gui:latest registry.nersc.gov/m558/superfacility/gui:$(date '+%y.%m')",
<         "docker push -a registry.nersc.gov/m558/superfacility/gui",
---
>         "docker tag gui:latest registry.nersc.gov/m558/superfacility/ml-training:latest",
>         "docker tag gui:latest registry.nersc.gov/m558/superfacility/ml-training:$(date '+%y.%m')",
>         "docker push -a registry.nersc.gov/m558/superfacility/ml-training",

[EZoni]

If it comes down to basically replacing dashboard with ml and gui with ml-training, I would try to make one single script with a command line option to decide which container image to publish, e.g.,

python publish_container.py --dashboard
python publish_container.py --ml-training

or, maybe more concise,

python publish_container.py --gui
python publish_container.py --ml

If you want, I could push some changes to this PR to implement this - let me know.

[ax3l]

How much of this code is similar to the one in publish_dashboard.py?

95% :D

[ax3l]

If you want, I could push some changes to this PR to implement this - let me know.

Ok sounds good. I can do it real quick.